What's New

What's shipping in CryptoComply, newest first. Each entry calls out new features, improvements, fixes, and anything to watch for.

A few acronyms used below: RFI (Request for Information), PEP (Politically Exposed Person), MCP (Model Context Protocol — the standard for connecting AI tools), KYC/AML (Know Your Customer / Anti-Money Laundering), and SSO (Single Sign-On).

8 October 2026 — A roomier Comply AI chat, a consistent look, and a clearer review pipeline

What's new

• Comply AI uses newer models. The assistant now runs on Claude Sonnet 5.5, with Claude Haiku 5.5 for quick tasks such as naming a chat, and each falls back to an earlier model if it is unavailable.
• Comply AI asks before changing a form draft. Saving, editing or replacing a form's draft through the assistant now shows what the draft will become and waits for your approval.

What's improved

• A roomier Comply AI chat. The chat page uses the full width of the screen, and each reply lists the steps it took, such as "Listed profiles", on one line you can open. Past chats open from the Conversation History button, which replaces the clear-chat button.
• Smaller text on phones. The assistant chat and every text field use a smaller size on a phone, and tapping a field no longer zooms the page in.
• A more consistent look. Notices, Show More buttons, tags, avatars, badges and rationale boxes look and behave the same on every page, and field names read the same everywhere, such as Date of Birth and Tax ID.
• Cleaner dashboard charts. The Alerts chart shows one stacked bar per month with round-number axis counts. The Review Pipeline chart no longer counts archived profiles, and every status with at least one profile keeps a segment wide enough to show its count, so a single escalated profile next to a thousand approved ones is easy to spot.

What's fixed

• Confirmation cards use labels. Claiming or assigning an alert, and deciding a Duplicate Submission alert, through Comply AI now show each change by its label, such as Assigned To.
• Invitation send counts agree. The result of sending invitations and the notification that follows report the same numbers of sent, skipped and prefilled invitations.
• Risk Engine with many custom thresholds. An organization with more than 25 custom thresholds saw the rest as Default and could not edit them. Every threshold now loads.
• Resolve Hit on a phone. The decision buttons are no longer squashed to the height of their text.
• Field comparison. Opening a field's "+N" other values opens one window instead of two.
• Tag names in Settings > Groups. Hovering a tag shows its full name.
• Periodic code security audit completed.

7 October 2026 — Forms reopen where you left off, and a more dependable Comply AI

What's improved

• Comply AI finds the right tool more often. You can ask in everyday words, such as wallet, customer, comment, label or dismiss, and look up a wallet by its address. When nothing fits your request, it says so instead of guessing.
• Comply AI only offers what your organization has. Knowledge base and adverse media actions are offered only to organizations that use those features.
• Comply AI explains how to get past a refusal. When it cannot do something, it says what would make it work, and it no longer retries the same failing request.

What's fixed

• Forms reopened partway through keep earlier answers. Someone who left a form and came back was taken straight to the page they had reached, with earlier pages blank, and could send it without noticing. The form now reopens on the first page that still needs an answer and will not send until every page is complete.
• Refused Comply AI actions show as failed. When you reopen a conversation, an action the assistant was not allowed to complete shows as failed, as it does live. Actions you approve no longer fail afterwards, since the preview now refuses anything the final step would refuse.
• Invitation steps are readable. Steps where Comply AI matches invitation recipients to profiles, or changes the profile an invitation is filled from, show a readable description.
• Alerts under review count as open. Alerts that are under review or escalated now count as open in Comply AI's analysis of addresses and profiles.
• Chainalysis screenshots on busy days. When several alerts arrived within seconds of each other, some Chainalysis panel screenshots failed. They are now taken one at a time, and a panel that does not load is tried once more.
• No more spinner after a release. The app no longer stays stuck on its loading spinner after a release. A browser already stuck needs one hard reload (Empty Cache and Hard Reload).

6 October 2026 — Duplicate submission alerts, prefilled invitations, and branded invitation emails

What's new

• Duplicate Submission alerts. When someone fills in an intake form and looks like a person or company you already have, their answers are held as a Duplicate Submission alert instead of creating a second profile, with a match score and a side-by-side comparison of every detail. Decide Same Person (Same Entity for a company), which copies the existing profile into the form's group with the new answers, or Different Person, the usual choice, which creates a new profile that starts Under Review. The form's numbers show how many submissions are waiting.
• Copies show where they came from. A profile copied by a Same Person decision shows the submission it came from, which details changed, and whether each document arrived with the submission or was copied over.
• Prefilled form invitations. Before an invitation-only form goes out, from the app or through Comply AI, you see which of your profiles each recipient matches by email and name and choose the one, if any, that fills in their form. The invitee finds their saved details already filled in, and their answers come back as a Duplicate Submission alert on that profile.
• Invitation matches in the list. The invitation list shows who each invitation matched and which profile fills in its form, and Change Profile swaps it until the invitee opens the link. Each bulk send records who sent it and how many were prefilled, and a profile's View Submissions lists its form invitations.
• Prefill on invitations, per custom field. A "Prefill on invitations" checkbox, ticked by default, sits on each custom question. Untick it and a prefilled invitation leaves that question blank.
• Comply AI handles duplicate submissions. It can walk you through a Duplicate Submission alert, show a profile's copies, and send prefilled invitations.
• Formatted email text. The form's Email tab has Bold and Link buttons instead of raw tags, and pasting keeps only the words. Links must start with https:// or mailto:.
• Name check in the Builder. A warning appears when a form asks for both Full Legal Name and First, Middle and Last Name, or for neither, with a button to remove either set.
• Repeated profiles across your organization. Comply AI can check profiles or outside people for repeats, showing who is the same subject, how their data changed and where it contradicts, matched on email, wallet, document number, name, phone or a copy link. A suspicious email sweep finds emails that look like a phone number, a generic first name on a company domain or another profile's full name.

What's improved

• Also In links say which group and which profile each one opens.
• Pop-up messages with a button, such as Undo, stay on screen for 15 seconds.
• Large invitation sends go out in smaller waves.

What's fixed

• Cloning a form starts from the source's published questions instead of its unpublished edits.
• Builder canvas. A page's description shows once, in line with its title, the form header is as wide as the pages below it, and Add a Question takes a light lime on hover. Button Text shows the colour the email actually uses.
• Alerts on a phone keep the same margins as the rest of the app, a window's title no longer runs under its close button, and a notice's icon stays centred beside its text.
• Screen readers announce the duplicate comparison's match levels, expandable sections and decision headings properly.

5 October 2026 — Review adverse media and watchlist hits on one screen, and branded invitation emails

What's new

• Review adverse media on one screen. The Adverse Media tab and the alert page show the findings as a list beside the one you are reviewing: its summary, why it matched, and each source page read right there from a saved copy, with the person's own details alongside for comparison and the decision buttons underneath. On the alert page you can claim, decide every finding, and close or escalate without going to the profile.
• A Watchlist Screening tab on the profile. Every watchlist hit a profile has raised, grouped by alert, can be decided as True Match, False Positive or Needs Info without opening the alert. The profile's reviewer can decide all of its hits; anyone else decides the hits of alerts they hold, or claims an unassigned one. The unresolved-hits count on a related party now opens this tab.
• Counterparty details. Each counterparty of a Chainalysis wallet screening shows its hops away and country, and appears as soon as the screening finishes. A sanctioned counterparty opens to show which jurisdictions designated it, since when, and the lists behind each. The counterparty filter also matches a country or designating jurisdiction.
• Branded invitation emails. The Email tab splits into Branding, shared by the invitation and its reminder (the form's or an uploaded logo, its width and alignment, and every colour with contrast warnings), and Content, written per email: button label, bullet and numbered lists, sign-off and help blocks, and your company's name, address and support contact. A phone-width preview shows both emails.
• Profile submissions. View Submissions in a profile's Actions menu lists the intake submissions behind it, as the subject or as a related party, and opens any one.
• Revoke an RFI. An RFI not yet answered can be revoked from its history with a rationale of at least 10 characters, kept in the audit trail. Its link stops working at once and the contact is not notified.
• Exports are recorded. The organization's activity log records who exported profiles, wallets or alerts, in which format and how many rows.

What's improved

• Notes are required to resolve a watchlist hit. Marking a hit as a true match or a false positive needs a note explaining the decision.
• Screening comparisons list the ten biggest counterparty changes first, with a Show All button.
• Adverse Media Search is shown in every organization and says it only opens a web search and is not screening.
• Dark mode uses one consistent palette, and status, risk and tag chips share one size and style.
• Accessibility and speed. Tables, the sidebar, the public form and buttons work with screen readers and the keyboard, each record's page names itself in the browser tab, and the app downloads about a fifth less code on first load.
• Saved copies of cited sources. More cited pages open as a saved copy, including pages the Internet Archive holds even when its quick lookup says it has none, and firewall refusal pages are no longer kept as the copy of a source.
• Sanctions lists on counterparties also include the lists Chainalysis reports designating the counterparty, and the reason Chainalysis gives for a wallet's risk shows as soon as the screening finishes.
• A profile's reviewer can rule on its adverse media findings, as with watchlist hits; closing or escalating the alert stays with whoever holds it.

What's fixed

• Counterparty and wallet totals no longer count a direct amount twice, and the profile PDF lists all watchlist hits instead of stopping at 25 per profile.
• Exports match the list. An export applies what is typed in the list's search box.
• UK/EU-only sanctions carve-out now also covers a wallet owned by an entity listed only by the UK or the EU.
• Escalation reasons are saved. The reason typed when escalating an alert was being dropped.
• Alert history wording. Decisions read, for example, "Alert decision changed from EDD Required to Risk Accepted".
• Public form and RFI links end on one branded page whether the form was sent, closed or cannot be opened, and an expired invitation link says it has expired.
• Send RFI and Actions appear only to members who can use them, and API keys load only for members allowed to view them.
• Invitation and RFI statuses stay put when the email sent after an answer or a revoke fails.
• Comply AI answers in the language you ask in, and confirming an action from a page button no longer fails.
• Transaction alerts from Slack. A forwarded client alert is handled as if the client had posted it, alerts missed during a brief outage are picked up within minutes, each new alert about a returning user opens its own alert and thread, and notes cover only what triggered the alert.
• Unpublished forms no longer hand out a link before publishing, and time zones show in date tooltips.
• Form style in the Builder draws in the form's own style, and Outlook shows the invitation button at full size and colour.

3 October 2026 — More attachment types

What's new

• New attachment types. Profile attachments can be filed as AML / Compliance Policy, Tax Registration, Financial Statements, Due Diligence Questionnaire or Contract / Agreement, and RFIs can ask for them.

2 October 2026 — Approval checks, Chainalysis screenshots, and intake form checks

What's new

• Approval blocked by open screening. A profile cannot be approved while it, or one of its wallets, has open alerts or undecided hits. A window lists each one with a link. Declining or archiving a profile closes its open alerts as cleared, recorded as the system.
• Approved customers filter. The alert list can show only the alerts of approved, active customers, with sanctions hits first.
• Chainalysis screenshot gallery. A wallet's Chainalysis capture shows captioned thumbnails, with each high-risk category opened onto its counterparties and sanctioned ones outlined in red; a viewer steps through them all.
• Intake form visibility checks. The form editor flags questions no one can see, required questions hidden from a profile type, questions shown to only one profile type and follow-ups shown without their condition. Publish stays disabled until the blocking ones are fixed, and an Apply Condition button adds the condition a follow-up needs. Question and page titles can be up to 512 characters.
• Offline and expired sessions. Going offline shows a notice and reloads when the connection returns, and an expired session sends you to sign in and back to the same page.

What's improved

• Faster pages. The app starts faster and loads lists, alert details and notifications with less work.
• Wallet risk history. Every change of a profile's risk rating or a wallet's risk level is in its history, and a screening that changes a wallet's risk is in the wallet's audit log.
• Consistent status colors across lists, filters and charts, and better text contrast, keyboard-openable table rows and Title Case buttons and labels.
• Identity document checks. Issue dates no longer count as a disagreement, a valid machine-readable zone is confirmed by its check digits, and an entity's proof of address and registers of shareholders or directors are compared with its registered details.

What's fixed

• Remember me. Signing in without it ends the session when you close the tab.
• Pages that cannot load show a message with a way back instead of loading forever or going blank.
• Pages, tabs and sidebar links follow the same permissions the server checks.
• Intake form saving keeps what the author wrote: a cleared description stays cleared, and wording a save cannot keep is refused with a reason. Conditions no answer can meet are refused.
• Comply AI step names name every action, such as closing, claiming or escalating an alert, in new and past conversations.
• Threshold changes. When new thresholds change a wallet's risk level, the change is recorded as the system and a rise raises an alert.
• Deleted attachments are also deleted from storage.
• Smaller fixes. Date filters include their last day, confirm buttons cannot send twice, a password change no longer signs you out, large organization cards show a sharp logo, the groups table shows tag names and dates, and exposure amounts match across wallet views and PDFs.
• Periodic code security audit completed.

What to watch

• API changes for integrations. Text fields that used to accept null now take "", and a list nested under a missing or another organization's record answers 404. Several MCP tools have clearer names, GET /api/orgs/ returns only the id, name and logos, and the risk-categories, profile risk-changes and inquiries routes are removed.

1 October 2026 — Copy profiles between groups, and ask before you rule on a finding

What's new

• Copy profiles into another group. Select profiles in the list and copy them into another group. Each copy is linked to its original as a duplicate, so it is always clear where it came from.
• Mark an adverse media finding Needs Info. Some findings cannot be ruled on without more from the person. Such a finding can now be marked Needs Info, with your reasoning. It stays at the top of the Adverse Media tab, marked as waiting on more information, with a Send RFI button that opens an information request for the details that would settle it; Comply AI can suggest what to ask. When the answer comes back, record the verdict on the same card. The alert cannot be closed while a finding is still waiting.

What's improved

• The profiles list keeps your place. Going back to the list keeps its search, filters and page, remembered separately for each organization.
• Comply AI answers in Spanish with the familiar "tú".
• Exposure chart legends shorten large amounts (for example $6.2K), and the total wraps cleanly on a phone.

What's fixed

• Form links with capital letters. A form whose link contains capital letters can be edited and saved from its Settings tab again, and the link opens however it is capitalized.

30 September 2026 — Exposure by direction, Arc addresses, and a Comply AI that knows more of your organization

What's new

• Exposure charted the way Chainalysis draws it. A wallet's exposure now shows as a receiving chart and a sending chart, each split into direct and indirect, with Chainalysis's own percentages. Clicking a category shows the counterparties behind it, and a table view lists the same figures. The two charts share one legend that fits on a phone. Wallets screened before this release keep the earlier charts until their next screening.
• See who is behind indirect exposure. The Indirect Exposure table now opens a category onto its counterparties, as the direct one does, with sanctioned counterparties listed first.
• Arc network. Arc addresses can be added and screened with Chainalysis or TRM from the platform, intake forms and RFIs, and show the Arc logo with a link to the Arc explorer.
• Comply AI knows more of your organization, and can do more with it. It can list RFIs and watchlist hits across the organization, filter alerts by who closed them and when, find profiles by their attachments or the RFIs sent to them, read your roles and optional features, and set who new profiles and alerts go to. It can also set a profile's risk rating and birth details, create an RFI as a link you share yourself, and match a list of names or emails to profiles already on file.

What's improved

• Clearer previews before Comply AI changes anything. Each field is shown by its label, and the preview says what the change will set off, such as screenings, alerts or emails. Every change is checked before you approve it, so an approval no longer fails afterwards.
• Dismissed submissions no longer count toward a form's metrics. They are left out of an intake form's total, success rate and response rate; the number dismissed still shows on its own.

What's fixed

• Comply AI chat behaves. The steps under a reply say what actually ran, a new conversation is titled at once, the conversation list keeps up as chats change, and long replies no longer drop text. A failed reply says why and offers a new chat, a chat that grows too long asks you to start a new one, and the floating buttons no longer cover the message box or the phone header.
• Unanswered profile fields stay empty. A new profile no longer gets "One-Time" activity or a "No" AML sanctions program unless someone answers them, and both show as unanswered on the profile and in its PDF report.
• Extra spaces in a name no longer make a watchlist hit read as a mismatch.
• Form styles are kept. A form styled before the Style tab existed now shows its saved colors and font there, and saving keeps them instead of resetting to the defaults. A form styled with Open Sans can be saved again.
• Testing a form. Jump to Page can leave a form's introduction and return to it, in the Style tab and in Test This Form, and Back to Form shows while a test is still being filled in.
• Smaller corrections: links to a record in another of your organizations open faster; saving a relationship says when the matching relationship on the other profile changed too; sending an invitation asks to confirm its CC recipients even while the form is still loading; long category names on the exposure charts are no longer cut off; and the test email and schedule review dialogs fill the screen on a phone.
• Periodic code security audit completed.

What to watch

• A few lookup tools retired for external AI tools connected over MCP. Their information now comes from the tools that use it. An AI assistant adapts on its own; a script or saved workflow that calls a retired tool by name needs updating. Comply AI inside the app is not affected.

29 September 2026 — See what a schedule change moves, and transaction alerts that explain themselves

What's new

• Review a screening schedule change before saving it. Changing a group's screening frequencies first shows how many profiles and addresses will move, with a list linking to each one.
• Archive a profile and end the relationship together. Archiving a profile offers to set its relationship to Inactive, and ending the relationship offers to archive it. Declining, archiving or ending a relationship now turns periodic screening off, and changing the profile's group no longer turns it back on.
• A revised alert decision is announced. Changing a closed alert's decision notifies the team with who changed it, from what to what, and why.
• Transaction alerts explain the recommendation and record the reply. The Recommendation tab shows why beside the note: the rule that applied and the entity behind each exposure. Sending something other than the recommended action asks for the reason, and once sent the alert shows the recommendation next to what was actually sent. The timeline names who sent each message and whether the suggested draft was edited.
• Work through every counterparty. A wallet now keeps every named counterparty its Chainalysis detail reports, not only the largest ones. On an address page and on a transaction alert's wallets, the counterparty table pages, sorts on every column and filters by name, category or sanctions list, with sanctioned and high-risk counterparties first. An address's PDF report lists sanctioned counterparties first, then the largest, and says how many more are in CryptoComply.
• Copy a profile's email with the button beside it on the profile overview.

What's improved

• Test emails go only to you. An intake form's test email reaches only the person sending it, and asks before also copying the form's CC addresses.
• A published form can change its defaults. Its group, profile type and blockchain can change after publishing, as long as its questions still fit them, and new submissions use them right away without publishing again.
• Clearer save messages on a profile. Saving a status names the new status and says when periodic screening stopped.
• Comply AI names statuses and risk levels the way the app does, and a bulk status change lists each profile's open hits, findings and alerts.
• Notes read cleanly. Lines wrap between words and bullets sit inside the text, and a note looks the same while it is edited as once it is saved. In the audit log a note's changes show formatted, with what an edit took out and put in marked.
• The alert side panel stays in view on large screens while you scroll through the evidence, so the note you are writing never scrolls away.
• Clearer refusals. Publishing a form names the question that stops it by its title, a refused status change lists where the profile can go instead, deleting a tag that is in use asks you to confirm, and a group name that clashes with another is caught before saving.

What's fixed

• Intake form drafts are never lost. When a draft changes in another tab or through Comply AI while you edit it, the editor says so and lets you keep your changes or load the new version. A save based on an older copy is refused, with who saved it and when.
• A manual screening keeps the schedule. It updates the Last Screening Date without moving the next periodic screening.
• Hit comparisons read like document checks. Accents, punctuation and word order no longer make a name a mismatch.
• RFI emails go to the contact, with any additional recipients copied.
• Bulk address screening screens an address listed twice only once, and reports a provider failure as failed.
• Emails fit a phone screen and read correctly as plain text, a single form invitation reads "The invitation was sent.", and an invitation to a recipient without a name no longer fails.
• Automatic RFI notes are logged as the system's, saying they were added automatically, instead of naming the officer as their author.
• Smaller corrections: profile overview values read in full (text fields widen while edited, dates show as dates, a long group name stays inside its card); avatar initials come from the first two words of a name and show when a photo cannot load; an open Compliance Health check and its table filter survive a reload, and the aging-alerts check is named "Alerts not closed after a week"; long values in Comply AI replies wrap inside the message; and when reviewing an RFI, Preview Client Form and the settings button sit in the email's header.
• Periodic code security audit completed.

28 September 2026 — Requests for information rebuilt as forms, and transaction alerts

What's new

• Requests for information, rebuilt as forms. Compose a request for the profile and each related party, preview it, and send it to many profiles at once, or straight from a Needs Info hit or a watchlist alert. Clients can save their reply and finish it later. What the reply brings fills the profile by itself and its documents are checked, leaving only changes to existing details for you to decide under Start Review.
• Transaction alerts. A new alert type for alerts raised by your own transaction-monitoring systems. Each wallet involved is screened and shown with its exposure and the Chainalysis view of it (click a view to enlarge it), the platform suggests a decision, and an analyst approves the reply before it goes back. Each alert links to the original message it came from. Transaction alerts are switched on per organization; ask your CryptoComply contact to connect yours.
• Advanced threshold options. An exposure threshold can set a de minimis amount and exclude exposure attributed only to certain sanctions lists. A wallet's screening result applies these options, so it no longer reads High for an exposure your thresholds have cleared.
• Exposure as donuts. A screening's exposure can be read as one donut per exposure type, next to the table.
• Rebuild a saved source page. Each page a screening hit cites is saved so it can still be read after the original moves or disappears. Some sites only build the part that matters once the page is open in a browser, so their saved copy arrived without it. A cited page whose saved copy is missing what the live page shows can now be rebuilt in a browser from a button beside it, and it keeps coming through correctly from then on. The original saved evidence is never changed.
• Electronically signed documents. A document signed as a .p7m file can be attached like any other, and shows as the document it carries.

What's improved

• Attachments are marked Reviewed instead of Approved. Attachments now read Pending, Reviewed or Rejected, and those already approved show as Reviewed.

What's fixed

• Client forms always show in light mode. An intake form or a request for information opened on a device in dark mode no longer shows a dark header over a light form.
• Comply AI waits for your approval. It shows a change and makes it only after you say yes.
• One click, one reminder. A double click on Send Reminder no longer emails the client twice.
• A file that cannot be opened is refused on submission, so the client is told right away and can attach another.
• Notes written by the platform are signed "System" instead of showing a blank or "Anonymous" author.
• Smaller corrections: leaving an untouched intake form no longer asks you to confirm; email previews are no longer cut off on the right; a screening hit without a match name shows the screened name instead of a blank; the organization's logo sits centered in information-request emails; and a form invitation sends when a name was typed and then cleared.

25 September 2026 — Every attached file arrives

What's fixed

• Attaching several files at once keeps all of them. Choosing more than one file at a time on an intake form kept only the first, and the form could be submitted while a file was still uploading. Every file is now kept, the form waits for uploads to finish, and a refused submission leaves the respondent on the form with their answers.
• Files named with two dots in a row attach normally. A submission whose file had a name like "Holdings Ltd..pdf" failed after it was sent, so no profile was created. A file that never finished uploading is now refused while the respondent is still on the form, with a message asking them to attach it again.
• A submission's files open again. Once a submission had been processed, its files showed as black or broken thumbnails and would not open. They now open from the profile they were filed on, and a file that is truly gone says so.
• Attached files are easy to tell apart. Files on an intake form show in a tidy grid under their full name and format (PDF, DOCX, XLSX and so on), in the form's own colors, with a clear notice when one cannot be shown.

23 September 2026 — Information requests ask only for what you requested

What's fixed

• A request asks only for what was requested. A request for documents alone showed the client the whole questionnaire. The client now sees only the fields, documents and questions chosen, each upload is filed under the document it answers, and a request that asks for nothing is refused.
• A passport no longer needs a back side. When an ID's back side was requested, a client holding a passport had to upload another file in its place. That slot is now optional.
• Every item in a reply is shown for review. The review of a reply listed only its first 25 items, while accepting everything also accepted the ones never shown.
• Accepted documents keep the type they were requested as, so a requested passport shows as identification rather than "Other".
• Owners and control persons accepted from a reply are screened, the same way as any other new profile.
• Sending an information request no longer asks for a reason. Where an organization requires a reason for changing a profile's review status, sending a request failed with "A rationale is required". It now sends in one step, and the profile moves to RFI Sent on its own.
• A refused profile edit says why. Saving a field or section that was refused showed nothing, and the change simply did not stick. The reason now appears on screen, and a blocked approval lists each related party still waiting on a decision.
• A profile's PDF report downloads again. Generating it failed every time with "Failed to generate PDF report". It now downloads, and a section left unchecked can no longer stop a report that leaves it out.
• Saved copies of cited pages keep their styling. Copies rebuilt from the web archive, or several from the same website at once, could come out plain and unstyled. They now come out whole.
• Periodic code security audit completed.

22 September 2026 — Rule on adverse media findings, and identity documents checked on arrival

What's new

• Rule on each adverse media finding. The Adverse Media tab now lists each warning the research turned up as its own card, the ones still waiting on a decision first. An analyst rules on each as a true match that matters, a true match that does not, or a different person, with the reasoning written alongside and the analyst's name and the date kept against the decision. A ruling can be revised later with a fresh explanation, and every ruling shows in the audit trail. The tab carries a count of what is still undecided, the way the Alerts tab does, and an adverse media alert cannot be closed while any finding is undecided. When the research runs again and returns the same finding, its ruling and reasoning are kept, so you review only what changed. Rulings are made by whoever holds the alert: if nobody does, the card offers to claim it. Comply AI can list findings and record rulings too.
• Identity documents are checked when a form arrives. When an intake form is submitted with an identity document, the platform reads it and compares every field with what the profile declares. Opening the document lays the field-by-field comparison beside it, moving to the page and passage each value was read from. Suggested next steps, such as an information request to send, are offers: nothing is applied on its own. The profile list can filter and sort on the outcome, so submissions whose documents disagree with them surface without opening each one.
• Read an attachment without leaving the list. An attachment now opens over the profile with its notes beside it, so a document can be read and annotated in one sitting. The arrow keys move through the rest, and a half-written note asks before moving on.

What's improved

• Reading a document no longer fills in the profile. It reports what the document says, quotes the passage each value came from, and leaves any change to a person.
• The activity log reads what the platform did. An entry made by an automated check shows each step, how long it took and its detail on request, instead of raw data.
• Integration accounts stay out of people pickers. Roles that exist only for internal integrations no longer appear in the role list, the permission matrix or invitation pickers, and their accounts no longer show in the member list as though they were colleagues. Nothing about who can do what changes.
• Saved copies of cited pages look more like the original, keeping more of its layout and images, and are ready sooner.
• The adverse media report opens on the findings that need attention, instead of keeping them behind a collapsed section.

What's fixed

• An adverse media alert can be raised again after it was closed. When later research found something new, reopening the alert failed and the research was lost. The alert now reopens with the new count and keeps its earlier closure on record.
• A document sent with a form can be read. Uploaded identity documents showed as tiny thumbnails under clipped names, or as a generic icon with a link that led nowhere. A submitted document now opens the way every other document on a profile does, where it can be enlarged and rotated.
• The invitation you edit is the one that arrives. Saving one part of an invitation email quietly reverted the others to the standard wording, and a body deliberately emptied to leave only the button was refilled. Each part is now kept as written.
• Identity documents are read more accurately. Nationalities written as passports state them (BRITISH CITIZEN, FILIPINO, Nederlandse) are recognised, a fuller legal name on the document than on the form counts as a match, and document types keep their capitals (National ID, not "national id").

18 September 2026 — Risk Engine shows which thresholds are yours

What's fixed

• Inherited exposure thresholds are marked as defaults again. The Default tag was missing, Custom Only left the list unchanged instead of showing only your own thresholds, and the bulk editor no longer warned that saving would create overrides for inherited rows. All three work again.
• A threshold can be added for a category still on its defaults. The Exposure Type list counted every category as already set, so the categories that most needed a threshold of your own could not be chosen. They are offered again.

17 September 2026 — Follow your system's theme, and open documents in place

What's new

• Light, Dark or System. The theme can now follow your computer's light or dark setting and change with it as you switch, without a reload; picking Light or Dark pins it. The choice is saved on each device, in the avatar menu and under Settings → Account.
• Open a document without leaving the page. The KYC documents on a watchlist alert and the file attached to an information request open over the page you are on, where a scan can be zoomed, turned upright and paged through with the arrow keys. Downloading is still one click away.

What's fixed

• Sorting alerts by date works, on the alerts list and on the Alerts tab of a profile.
• Documents attached while creating a profile are saved with it, instead of being dropped without an error.
• An intake form set to Entity is built as an entity form, rather than laid out with individual fields.
• Filtering API keys by type works, instead of returning the full list whichever type you picked.
• Organization settings no longer offers an Open Sanctions key, a field whose value was never used.

What to watch

• Saved column layouts on the alerts and profile lists are rebuilt once. A layout you had saved is rebuilt the first time you open those lists, and saves normally after that.
• Exposure-threshold spreadsheets exported before this release need re-exporting. Their column names changed, so export once more to get a file that can be imported again.
• An unfinished new-profile draft from before this release is discarded, so you start on a clean form rather than one whose fields no longer line up.

14 September 2026 — Wallets and profiles added from Slack screen right away

What's fixed

• Anything created from Slack is screened immediately. A wallet or profile created through the Slack assistant was saved, but its first screening never started, so it sat with no risk rating, and the assistant reported an error even though the record existed. Wallets added from Slack before this fix need a one-off screening from the web app, or a "re-screen" request in the channel.

11 September 2026 — Weekly and monthly screenings run on the day they are due

What's fixed

• Screenings run on the day they are due. A weekly screening in practice ran every eight days and a monthly one every thirty-one, and in between the profile showed as overdue for a whole day. Schedules now count calendar days, so a weekly screening happens on the same weekday every week.
• Live updates keep arriving after a quiet half hour. A tab left idle for more than thirty minutes could stop receiving notifications and screening progress until reloaded. It now recovers on its own.
• Every list of members reads the same way. The reviewer and actor filters list members alphabetically, with their photo and name, like the assignment pickers.
• Automatic entries in the activity log show their reason.

10 September 2026 — The activity log says what started a screening

What's improved

• Automatic screenings are recorded as the system's. The screening that runs when a profile or address is created was logged as one its creator had asked for. It now reads as an initial, automatic screening by System, and the same applies to the adverse media check. A screening you request yourself still shows your name.
• A default reviewer assignment explains itself. When the organization's default-reviewer setting puts a new profile on someone's desk, the activity log names that setting.

9 September 2026 — Settings and filters that match your role

What's fixed

• Risk Engine appears for the roles that can open it, and opens on the tab that is yours, instead of leading some members to an access-denied page.
• Filters and fields appear only for the roles that can use them. The reviewer and group filters on the profile, alert and Inbox lists, the group field on an address, and the exposure-threshold dialog failed for members whose role does not cover them.
• Default assignments are visible to the whole team. The tab under Settings → Users showed an error to everyone but administrators. Any member can now read it; changing a default still takes an administrator.
• Periodic code security audit completed.

8 September 2026 — Editing notes works again

What's fixed

• Edit Note opens again. Clicking Edit Note on a profile, an address or one of their attachments opened nothing. Notes can be edited everywhere they appear.

5 September 2026 — A faster Comply AI that no longer gets stuck

What's improved

• Comply AI answers faster. When a question needs several lookups it makes them at once rather than one after another, and a long answer ends with what it has instead of spinning.
• A conversation picked up days later checks again. Comply AI says how old its earlier lookups are and refreshes them before relying on them, and it knows today's date when you mention a month.
• Proposed changes are checked when they are proposed. Whether you are allowed to make a change is answered as soon as Comply AI proposes it, rather than after you approve it, and you have five minutes to review a proposal before it expires.

What's fixed

• Pressing Stop no longer breaks a conversation. Stopping mid-answer left that conversation unable to reply at all. Those conversations work again on their next message.

4 September 2026 — Review work has an owner, and an Inbox to find it in

What's new

• Profiles and alerts now name a reviewer, and every reviewer has an Inbox. Work can be claimed, handed to a colleague, or released back to the pool, with a written reason recorded each time. The new Inbox, directly below Dashboard, is one queue of everything assigned to you — profiles and alerts together — and an organization can name a default reviewer so new work arrives with a name on it already. The same assign control and dialog appear everywhere: the list pages, a row menu, and the profile overview.
• Invite someone who already uses CryptoComply. An address that already has an account can now be invited to another organization like any other. The invitation opens on a page that simply asks them to join, instead of a sign-up form for an account they already own — no second account, no second password. Signing in brings them back to the invitation rather than dropping them on the dashboard.
• The member list says when each person was last active, next to when they joined, so you can tell who still uses their access from who merely still has it. Activity is counted per organization, so one account never reveals what its owner does in another.

What's improved

• The member list shows people instead of three columns of name. Email, first name and last name were three columns saying one thing, and are now one. "Last Login" is now "Last Sign-in" — staying signed in means most people rarely sign in again, so that date said little about whether someone is still around.
• Inviting people no longer breaks on a pasted list. Addresses are read out of whatever you paste, a repeat is marked instead of silently disappearing, and anything that is not a valid address is flagged in place rather than sent and bounced back. Clicking an entry to correct a typo keeps its place in the list, and a space now ends an address the way a comma already did.
• Making a role no longer costs you the invite. Needing a role that did not exist yet meant leaving the invite and losing the addresses you had typed. You can create one from the invite, and it comes back selected.
• Accepting an invitation puts you in that organization, instead of handing you the organization picker one screen after welcoming you to a particular one.
• A rejected invitation points at the address it is about. Sending a batch where one person is already on the team reported a list of messages under the field, leaving you to work out which address each one meant.

What's fixed

• Deactivating an account now ends every way it reaches the product. A deactivated person could still receive password-reset mail, still lend their role to a linked Slack account, and still block their own re-invitation.
• Choosing a group brings up the rest of the new-profile form. Picking a group left the page showing nothing but the internal label and group boxes, so there was no way to fill anything in.
• Sorting the member list works — every column offered to sort and none of them did.
• The screening card no longer contradicts itself. A hit could say a name matched while the field comparison underneath showed it did not. The provider's verdict is now labelled as the provider's, separate from the comparison, and each part explains what it is claiming.
• Notification settings say which alerts cannot be switched off. Events that are always sent gave no sign of it, so turning them off appeared to do nothing.
• Smaller corrections: an organization with no logo shows its initials instead of a broken-image icon; a list steps back a page rather than stranding you on an empty one; a note nobody typed in counts as empty; the profile overview card reads Frequency rather than Screening Frequency; a member's tooltip points at their name instead of covering the whole row; and a locked menu item now explains itself to the keyboard, not only to the mouse.

3 September 2026 — Organization logos look like the file you uploaded

What's improved

• Every surface that takes an organization logo now says what to upload. Nothing said what a good logo looks like, so a logo too small for the organization switcher was only discovered by looking at it afterwards. The app now asks for a square image of at least 400×400 — JPG, PNG or WebP, under 5MB — before you pick one. Nothing that was accepted before is rejected now; a smaller logo is still your call.

What's fixed

• Logos are sharp in the organization switcher. The large and medium tiles asked for the smallest stored copy — a hundred pixels square — and stretched it to four times that size, so every logo looked soft however good the file behind it was.
• Uploading a logo keeps the resolution of the file you chose. The crop step saved the picture at whatever size it happened to appear on screen inside the dialog, so a large, detailed logo arrived a few hundred pixels wide and the result depended on the size of your browser window. Logos uploaded before this release keep the resolution they were saved at and need uploading again to benefit.
• A logo with a transparent background no longer sits in a grey box in the organization list.

2 September 2026 — The intake form editor becomes the form

What's new

• Build the intake form you actually need, and see it as you build it. The editor showed a list of fields on one side and a preview on the other, so you were reading two versions of a form and trusting they agreed — and the form it could describe was narrower than the form it had to serve. The middle of the editor is now the form itself: what you see is what the person filling it in sees, clicked and typed over in place. Pages, panels, repeating sections for wallets and related parties, document uploads, consents and paragraphs are placed on it directly, with rules that hide a question until an earlier answer calls for it. A form can ask a person and a company different things, and word the same question differently for each. Editing a published form touches a draft, so a half-finished change never reaches anyone until you publish it. Testing a form is a step inside the edit loop rather than a dead end — finish a test and you come back to the editor, ready to publish.
• Ask your own questions, including consents. An organization can define its own question without an administrator, and one of the kinds it can define is a consent: a block of copy with a checkbox per statement. What a consent leaves on the profile is the record of what was shown and how each statement was answered, declines included — which is what an audit later needs to see. A profile now shows each statement with its answer instead of collapsing the whole thing into one value. Source of funds joins the answer lists the platform serves.
• Every screen can be linked to exactly as you left it. A filtered table, an open side panel, the tab you were on, an expanded card, a step in an import — all of it now lives in the address bar. Send someone the link and they see what you were looking at. The back button goes back to what you were doing rather than out of the page, and a reload keeps your place instead of resetting to the top.
• Open CryptoComply AI from a link. Adding ?ai=true to any page opens the assistant panel there, so a link can hand someone both the screen and the conversation about it.
• CryptoComply AI can tidy tags, read a compliance-health check, and trace who is connected to whom. It can now rename and retire tags rather than only create them, report what needs attention across the book, show who touched an alert and what they wrote down when they closed it, and walk the whole web of relationships around a profile in one step. Closing a batch of alerts now says up front how many will actually close and why the rest will not.
• One result email for every bulk operation. Uploads, invitations and reminders each reported back in their own email, worded differently and carrying different detail. They are now a single email, and when a run has failures they are grouped by the column that caused them — so you are told which column to go fix rather than only how many rows did not make it.

What's improved

• An intake form stays live when someone else changes it. A form edited by another author, by Comply AI, or by the system now reaches your open page without a reload. If your canvas is untouched, the incoming change is adopted; if you have unsaved edits, your work stands, saving is held, and you are offered a reload rather than one of you quietly overwriting the other.
• Consistent capitalization across the product. Headings, buttons, labels and table columns were a mix of styles and now read the same way everywhere.
• Risk Engine settings are shown only to people whose role lets them see the organization's settings.

What's fixed

• Having the app open in several tabs no longer signs you out. Tabs raced each other to renew the same session, and the loser was signed out. Renewal now happens once, under a lock the whole browser respects.
• Switching organizations no longer flickers back to the previous one, and the dashboard no longer flashes "Unable to load dashboard data" with a Retry button that did nothing while your organization was still loading.
• A form that could not accept an answer is refused before it is published, not after. A form asking a person and a company different things could be published in two states that looked fine and then failed every response from one of the two.
• A response that fails to process now says why, naming the answer that could not be stored, instead of sitting in the list marked failed with an empty reason.
• Inviting someone to a form that has not been published is refused with the reason, before any email is queued, rather than sending them to a form that cannot be served.
• The buttons in emails now go where they say. The result email for every bulk run drew a button that led nowhere, and "manage your preferences" pointed at a settings page most people cannot open.
• Attaching a document no longer destroys the file it is saving. The original was deleted in the same step as the copy, so if anything went wrong afterwards the retry found nothing left to read and the upload could never succeed.
• Reading an identity document works in organizations that require a reason for every change. An uploaded document fills in profile fields by itself but recorded no reason, so the whole reading was refused and nothing was filled in.
• A table's page and filters no longer leak between sibling tabs. Paging the Alerts tab of a profile applied that page number to Blockchain Addresses, and paging one Compliance Health card paged every other card.
• CryptoComply AI can build and edit forms. Every change it tried to make to a form failed and came back as a generic fault, so it retried the same thing. When one of its actions fails it is now told what failed and whether it is worth trying again.

1 September 2026 — More Slack channels, and citations that render

What's new

• A client can connect more than one Slack channel. A client running several rooms — a deal desk, an escalations channel, a shared channel per counterparty — can now have all of them act on that client's data. An organization was previously limited to exactly one channel. A channel still belongs to exactly one organization, so there is never a question of which client a message belongs to.

What's improved

• Adverse media stops waiting on watchlist screening. The watchlist provider permits one screening at a time per account, so screenings necessarily queue. Adverse-media checks depend on a different provider entirely and were queuing in the same line for no reason. Results now arrive sooner when screenings are busy.
• Readable copies of cited sources arrive instead of stalling. A page borrowing a stylesheet from a site that has since gone offline left the render waiting about a minute on a request that would never be answered, then gave up with nothing to show; a public web archive that was throttling us could add three minutes and buy nothing. Those pages now render in seconds with anything unreachable left out, and citations that had no readable copy for these reasons are rebuilt on the next pass.
• A copy that rendered imperfectly is marked as such and re-rendered automatically when the machines are quiet, rather than being kept silently as the final version.

26 August 2026 — Act on alerts from Slack

What's new

• Analysts can act on alerts from Slack, not just read them. Asked to assign an alert and resolve a hit as a false positive, the assistant could look everything up and then had to send the analyst to the web app. It can now claim and assign alerts, resolve screening hits, close alerts and add notes, as the person who asked. What they may do is exactly what their role allows: an analyst with read-only access still cannot close anything. Work done this way is attributed to them, so an alert they resolve from Slack is theirs in the record, with the audit trail also naming the assistant that carried it out.

What's fixed

• Having the app open in more than one tab no longer signs you out. Each tab renewed its session on its own, and the platform retires a session as soon as one of them renews it — so whichever tab was second found its session already gone, several times an hour.
• A citation is never labelled with an error page's title. When the copy first captured of a cited page turned out to be an access-denied wall, a geo-block notice or a parked domain, the citation kept that page's title even after the genuine page was found in a public archive.
• Address enrichment waits for the panel instead of giving up in minutes. The named counterparties and money flows fetched after an address is screened are only available once the vendor panel has caught up, which can take several minutes. An address the panel still has nothing for is now revisited by the periodic backfill instead of being recorded as unavailable.
• Adverse-media analysis no longer fails when a subject has a lot of press. For a person or company with many articles the write-up could exceed the length allowed, and the whole screening failed.

24 August 2026 — Office documents, and screenings named consistently

What's new

• Attach spreadsheets, Word files and CSVs. The file picker now offers Excel, Word, PowerPoint and CSV alongside PDFs and images, so the documents due diligence actually arrives as can be attached directly.

What's improved

• Screenings and the hits inside them are named consistently. What used to read as an "inquiry" is now a screening, and each watchlist, PEP or adverse-media match is a screening hit.
• An audit-log entry made through Slack names the person it was done for. Rows performed via the CryptoComply Slack bot showed a bare bot; they now show the person the bot acted for, with the bot's mark overlapping their avatar. The actor filter gains a bot option, and CSV, JSON and PDF exports of the trail carry the same attribution.
• Alerts with many hits open faster. An alert can hold hundreds of matches, and the full vendor detail for every one was loaded to draw a list where only one card is open at a time. That detail is now fetched when you expand a card.
• Screening frequencies and relationship types come from the platform rather than being written into the app, where they could drift from what the platform actually accepts.

What's fixed

• Periodic code security audit completed.
• An already-submitted form link shows the right screen instead of an error when you reopen a link you had already completed.
• The relationship chart draws every connection — nodes were being dropped, so the chart could show less than the profile actually holds.
• Withdrawing cookie consent now stops analytics without needing a page reload, and notification categories read the same way everywhere.

19 August 2026 — Related parties get the full profile treatment

What's new

• Adding a related party now offers everything creating a profile does. The Add Related Party dialog used to ask for a fixed handful of identity details; it now shows the same sections your group defines — including your organization's own custom fields — and lets you tag the party, write its first internal note, and attach blockchain addresses while creating it, all in one step. Which fields appear (and which are required) follow the group's configuration, so it no longer insists on details you don't collect.

What's fixed

• The Add Related Party dialog fits on the screen. The form used to run past the bottom of the window with its save button out of sight; it now scrolls inside the dialog with the buttons always in view, on phones too.
• A related party that can't be saved says why, instead of a generic "please try again," and the relation type and person/company choice you made are carried into the form — switching between person and company keeps what you've already typed.
• Recording a beneficial owner, control person, or parent for the first time no longer asks for a rationale. In organizations that require a rationale for profile changes, first-time entry of these details was wrongly rejected as an edit; recording new information is now accepted, while changing or clearing an existing value still requires a rationale.

11 August 2026 — Reliable updates and clearer intake questions

What's fixed

• The app no longer gets stuck refreshing after an update. When a new version was deployed while you had the app open, the automatic reload could keep serving the old copy and spin in an endless refresh; the reload now fetches the fresh version, and a Retry screen appears in the rare case a stubborn cache serves the old one twice.
• Faster repeat visits. The app's unchanged files are cached properly again, so returning to the app re-downloads only what actually changed.
• A form open to both people and companies asks each for its own name. On a form where the applicant chooses individual or company, the name question is now worded for that choice — a person is asked for their full legal name, a company for the entity's legal name — instead of one wording that could leave companies unasked (and their submissions unable to become a profile).

10 August 2026 — Revisit decisions, step through alerts, and hide empty fields

What's new

• Change a decision you already recorded on a hit. A resolved hit is no longer a dead end — its assignee can reopen the decision, with the existing rationale carried over to amend (the reasoning has to actually change to be accepted). If the alert was already closed, you're asked whether its disposition still stands. The same revision is available through Comply AI.
• Step through alerts without going back to the list. An alert opened from a list now has previous/next controls in its header, the way profiles and blockchain addresses already did.
• Hide the fields a profile never filled in. A switch drops every empty field across a profile's information tab (including sections left with nothing to show); the choice is remembered in your browser and applies while editing.
• Show a hidden field to the whole group in one click. The warning marker on a field that's hidden by the group but holds data is now a button that names the field and group and asks to confirm, instead of sending you to settings.
• Profiles open the way you left them — a collapsed overview stays collapsed on the next profile you open (following a link to a section still expands it).
• See which role you're signed in with. The sidebar user card and menu now show your role in the current organization, and update when you switch organizations.

What's improved

• Redesigned watchlist hit card. The side-by-side comparison of your customer against the listed party now leads the card at full width, with the vendor's details, narrative, and citations below; the card shows every identifier the record carries (all passports and listing numbers, not just the first) and links out to the vendor's own record.

What's fixed

• The audit log on an alert opens again (it had shown "Something went wrong"); the notification-subscription screens name their events and categories again; ⌘/Ctrl-click and "Open in new tab" work on an alert's Related cards; the profile and blockchain-address action menus match the rest of the app and are readable in dark mode; plus assorted section and menu polish.
• Signing in still works when rate-limiting is briefly unavailable — sign-in, session renewal, password reset, and public intake forms no longer fail if the rate-limit service is momentarily unreachable.

7 August 2026 — Explained matches, durable source citations, and a leaner vendor lineup

What's new

• Match explanations on screening hits. A hit now shows why its score is what it is — the exact listed term that matched (flagged when it's an alias/AKA rather than the primary name) and a field-by-field verdict (name, date of birth, country, gender: matched or not) — so an ambiguous score can be adjudicated at a glance. Works retroactively for already-stored World-Check hits.
• Additional details on hits. The vendor's full narrative on a listed party — biography, identification details, reports, and per-regime sanction notes — is captured with each hit and shown as structured, collapsible sections, instead of something to look up in the provider's own console.
• Durable source citations. Every source cited by a hit is resolved and self-hosted: each link shows its status — Live, Wayback (a date-matched web-archive snapshot), Archived copy only, or Unavailable — and "View source" opens a self-contained, sandboxed copy that works even when the original site is gone (files like PDFs and spreadsheets are offered as a download). A live page that has drifted from what was cited is flagged.

What's changed

• A leaner screening vendor lineup. Castellum, OpenSanctions, and Hypernative have been retired: watchlist screening now runs exclusively through World-Check, and blockchain-address screening through Chainalysis and TRM. Organizations on a retired vendor were moved over automatically, and historical results, hits, alerts, and resolutions are unaffected — only running new screenings against those vendors is gone.

What's improved

• Attachment filtering, rebuilt. The attachments panel now filters by source and document type through two dropdowns (each option showing how many attachments it would leave, and options that would leave none hidden), plus a search box across attachment names, filenames, and notes — replacing the row of chips that grew with every document type. Attachment cards are calmer, with badges and the type picker sized to the panel.
• A stronger Comply AI. The assistant moved to Anthropic's latest Sonnet model, better at multi-step research across your data, at no change in cost. Asking for your riskiest wallets or most recently changed profiles now works (previously it could only sort by name or creation date).
• Scheduled screening failures are visible. When a scheduled watchlist screening can't complete, the profile's schedule now records what went wrong and when, clearing automatically on the next successful run.

What's fixed

• Document category names read properly again ("Screening & Monitoring Reports" no longer collapses to "Screening Monitoring Reports"); the attachment type list stays inside the panel; related parties load in a single request and show the profile's whole related network; navigating away from a still-loading page no longer shows an error; member pickers list every member (not just the first 25) and say so when a list fails to load; blockchain-address pages no longer flash "No data available" before results arrive; and a fully reviewed request for information now shows as Closed instead of staying Responded.

21 July 2026 — Attachment provenance and blockchain-address screening detail

What's new

• See where every attachment came from. Each attachment now shows how it reached the platform — uploaded by your team, submitted through an intake form, or provided in an RFI response — along with who uploaded it and when.
• Redesigned attachments panel. Attachment cards are more compact and scannable; document type and review status are edited inline (status shown as a colored badge, with a confirmation before a change takes effect); attachments can be filtered by source and by type; and the type list now covers screening-report and corporate-document types that previously had to be filed as "Other".
• Far more detail on a screened blockchain address. The address page now surfaces an Address Metrics section (incoming/outgoing attributed-exposure USD, first/last-active, transfers in/out), a plain-language risk reason, and address-type / VASP classification chips with tooltips. Sanctions-list attribution appears as badges on sanctions rows and associations (e.g. OFAC SDN, UK OFSI); a match whose list can't be identified is called out as a distinct "Review required" state rather than shown as clean. A new Counterparties table lists the named entities an address transacted with, most material first, with per-direction USD and their own sanctions badges. All of this is mirrored in the PDF screening report and the screening-comparison view (including a counterparty diff between two screenings).
• Counterparty categories drill down. Each Counterparty Exposure category row now expands to reveal the named counterparties in it, with sanctions badges and sent/received amounts, on both desktop and mobile.
• Open any row in a new browser tab. Hold ⌘/Ctrl (or middle-click) on a row in any list — profiles, related parties, addresses, alerts, intake forms, groups, compliance-health results — or on a profile's tabs, to open it in a new tab without losing your place.

What's improved

• Longer notes. Notes now accept up to 50,000 characters, so long compliance narratives no longer need to be split across several notes.
• Tidier tables. Status and type badges within a column now share a single width, so columns line up cleanly.
• Where an address's identity lives. The address-type and VASP chips moved into the Address Identification section (the risk reason stays with Risk Analysis), so identity and risk are cleanly separated on screen, in the comparison view, and in the PDF.

What's fixed

• Periodic code security audit completed.

16 July 2026 — Open a submission from the forms page

What's new

• View a submission without leaving the forms page. On an intake form's Submissions tab, click any submission to open the original form exactly as the submitter saw it, with their answers and consents filled in — the same read-only view already used on a profile's page.

15 July 2026 — Guided bulk-import wizard

What's new

• Import from a CSV in guided steps. Import profiles or blockchain addresses from a CSV: pick the group, download a template that matches its fields, upload, then review your data in an editable spreadsheet. Each column maps from its own header, flagged cells are highlighted (red for errors, amber for warnings) with the reason on hover, and the row number turns red so a bad row stands out. Fix cells inline and re-run the checks without re-uploading, and open a reference of every accepted field and value from the sheet.

13 July 2026 — Mobile support and dark mode

What's new

• Full mobile support. Every page, table, form, and modal adapts to small screens, with a bottom navigation bar, full-screen dialogs, and readable text sizes throughout — across the dashboard, profiles, addresses, alerts, settings, notifications, the AI chat, intake forms and their builder, and the public intake/RFI forms your clients fill in on their phones. Buttons are easier to tap, and typing no longer triggers unwanted zoom on iPhones.
• Dark mode. A dark theme spans every authenticated screen, toggleable from the sidebar (or the mobile menu). It's driven by centralized theme tokens, so surfaces, text, borders, and the lime brand accent stay consistent everywhere. (The intake-form survey styling keeps its own theme.)

10 July 2026 — Phone-photo uploads and consistent labels

What's new

• Phone-photo uploads just work. The attachment and RFI file pickers now accept HEIC/HEIF, TIFF, WebP, BMP, and AVIF alongside PDF/PNG/JPEG, so iPhone photos upload without conversion (the platform transcodes them to a viewable image), even when the browser reports no file type.

What's improved

• Country and choice labels resolve consistently. Gender, ID type, custom dropdowns, and country fields now render their readable name from a shared catalog everywhere they appear — in read mode, in the PDF report, and in the watchlist side panel — instead of depending on a label travelling next to each value.
• Full custom-field types and conditional logic in forms. The form renderer now supports the complete custom-field type set and the current show/hide condition rules across profile, RFI, and intake-form surfaces.

9 July 2026 — Custom RFI questions, follow-ups, and reminders

What's new

• Custom questions on an RFI. The compose modal gains a Custom Questions builder — type free-text questions (up to 25, add/remove/reorder) that go out alongside the fields and documents you request. An RFI can be question-only. The public form renders them as a required "Additional Questions" section, and answers come back in the response-review drawer with each question as the heading; accepting one records the answer on the audit trail without writing anything to the profile.
• Follow-up RFIs. A responded RFI gets a Send follow-up action that reopens the compose modal in follow-up mode; the new RFI is linked to the original and carries a "Follow-up" badge. Follow-ups are typically question-only.
• RFI reminders. The profile's Requests-for-Information drawer now has a Send reminder action on live RFIs, a reminder count and last-reminded timestamp, and a View reminder email preview. Each RFI row also lists exactly what was requested — the fields and documents — as chips.

What's improved

• Forms can ask a question in plain language. A field now carries a separate recipient-facing question and helper text, distinct from the internal audit label — so a field whose audit noun reads awkwardly ("U.S. citizen") can ask it properly ("Are you a U.S. citizen?") on the form without changing what staff and the audit trail see.

What's fixed

• Faithful RFI email preview. The sent-email preview now renders the email exactly as delivered, instead of clipping content, and references the actual "Provide Information" button.
• RFI checkbox alignment. A checkbox for a long, wrapping question label now sits flush with the first line instead of centering against the whole block.

8 July 2026 — Retry when a query fails

What's new

• Inline retry instead of empty sections. List tables, profile and alert detail views, blockchain-address lookups, intake/RFI drawers and summaries, and the settings group/risk-engine panels now show an inline error state with a Retry button when a request fails, instead of rendering an empty or stale section.
• Guards against losing edits. Closing or navigating away from an edited role or account form now prompts before discarding changes.

What's improved

• Faster and steadier. Several screens load faster, and modal dialogs now share consistent focus, dismissal, and scroll behavior.

What's fixed

• A crash in a banner outside the main app no longer takes down the whole app shell, and profile-screening schedule-expiry timers no longer leak across schedule updates.

6 July 2026 — Screening configuration shows correctly again

What's fixed

• Organization settings no longer show empty screening config. The Screening Configuration card could show empty created/updated dates and "Not configured" for services that were in fact configured, depending on which screen loaded first. It now always reflects the real configuration.

5 July 2026 — Intake header band renders reliably

What's fixed

• Header band no longer paints white intermittently. A timing- and stylesheet-order issue could leave an intake form's header band white until a refresh — hiding a logo with light elements. The band and its description color now render reliably on every load.

4 July 2026 — Email theming controls

What's new

• Email header-band, footer, and button-text colors. The intake-form Email editor's Colors section gains Header Band (a band behind the logo), Footer, and Button Text pickers, alongside Background / Card / Accent. Unset, they render exactly as before; set a brand color (e.g. navy) so a logo with light elements reads on it, and button text stays legible on any accent.

What's removed

• Organization brand-logo upload. Org settings no longer upload or show a "Brand Logo" — the email header logo is now configured per form instead.

What's fixed

• The intake header band no longer flashes white before the theme applies.

3 July 2026 — Content & field blocks, consent pages, and intake forms for everyone

What's new

• Content and field blocks. Add display-only content blocks (rich HTML/text) and generic fields (short/long text, checkbox, choice, number, date, email) that are recorded on the submission only — never written to the profile — for waivers, signatures, acknowledgements, and extra consents. A page can be marked a consent page, so its required checkboxes act as the consent acknowledgements.
• Editable landing group in Settings. The editor's Settings tab shows the form's default group and lets you reassign it; changing any submission-level default (group, profile type, or blockchain) opens a confirmation dialog explaining the effect first.

What's improved

• Intake forms for everyone. Intake forms and publishing are now available to every organization — the feature flags and their gating were removed.
• Editor polish. Field types read as human labels (Short text, Single choice, …), the Question Settings panel is wider, and conditions stack in cleaner cards.
• Readable legal and consent text. Consent, legal, and privacy blocks render with document-like styling — sensible heading sizes, paragraph spacing, and indented lists — and authored numbered lists render as legal-document enumerations with hanging indents.
• Optional/clearable defaults. A wallet section's blockchain falls back to the form's default, and the default profile type and blockchain can be cleared so the form asks the respondent instead.

What's fixed

• Phone fields no longer crash the profile page. Validating a phone field could replace the whole profile view with "Something went wrong"; it now validates normally.
• Readable header description on a colored band, and the read-only submission view now renders markdown/HTML labels (e.g. a consent acknowledgement) instead of raw tags.

2 July 2026 — Text-field options and per-page preview

What's new

• Short or long text, per question. Free-text questions have a Multi-line toggle in the editor — turn it off for a single-line input, on for a text box — available on every free-text field.
• Preview a single page. Each page in the builder has a Preview this page button that opens the preview jumped straight to that page, with a Back to builder link.
• Require a repeated section. Repeated sections (e.g. Beneficial Owners) now have a Required toggle — when on, the submitter must add at least one entry before continuing.

What's improved

• Longer consent text. The consent acknowledgement editor now accepts long legal notices (up to 100,000 characters).

What's fixed

• On dark form themes, multi-line text boxes and the date-picker icon stay legible; page/section titles containing & render correctly; and returning from a page preview opens the builder at the first page.

1 July 2026 — Custom form links, group selection, and custom themes

What's new

• Custom form links. An intake form's link can be given a readable custom address (e.g. celo-tip-intake) from Settings, copyable in one click, with a live preview of the URL.
• Choose which groups a form offers. Uncheck any group (e.g. internal or operational ones) to hide it from applicants.
• Default profile type and blockchain. Choose the default profile type and blockchain when creating any intake form, and change them later from Settings.

What's improved

• Dark and custom form themes. Intake forms render correctly on any color palette — descriptions, inputs, and borders adapt instead of staying light and unreadable.
• Faster page loads, especially the profile and blockchain-address detail pages, and a smoother form builder.
• Cleaner test-submission result — what a submission would create is shown as tidy, sectioned cards (profile, related parties, wallets, attachments).
• Page-specific browser tab titles. The browser tab now shows the page you're on instead of a single static title, so bookmarks are meaningful and multiple open tabs are easy to tell apart.

What's fixed

• Condition guardrails in the builder. The profile-type and group selectors can't be hidden, made conditional, or marked optional; a condition can only reference an earlier question; and removing a question clears any condition that referenced it.

30 June 2026 — Test an intake form before publishing

What's new

• Test this form. A "Test this form" action opens a form exactly as a respondent sees it and, on submit, shows a report of what a real submission would create — profile, fields, related parties, wallets/addresses — without saving anything. It runs against your current draft, so you can test at any point while building.
• Send test email. Send the real invitation email to your own inbox to check the template and delivery.

What's improved

• Optional group for address forms, since address forms create standalone addresses, and the form's header actions are grouped under a single Actions menu.

What's fixed

• Faithful rich text on public forms. Titles, descriptions, and consent text render their formatting and special characters (e.g. &) instead of raw HTML, and a wide header logo is centered and height-capped.
• In the builder, the profile-type and group selector cards show their question instead of the raw field key.

28 June 2026 — Intake form condition builder

What's new

• Visual condition builder. Add show/require rules to questions and pages from a visual builder — pick the source question, operator, and values — plus an "Applies to" control to target individual vs. business respondents or specific groups. An "Add Identification Page" shortcut sets up the profile-type and group questions, and the field picker groups fields by branch.

25 June 2026 — Blockchain-address attachments

What's new

• Attachments on a blockchain address. A blockchain address now has its own Attachments section, just like profiles — upload supporting documents (PDFs or images), classify each by type (Screening Report, Transaction Record, Ownership Verification, or Other), and keep per-attachment notes, each gated by your role's permissions.

What's improved

• Shared attachments component. Profile and address attachments now run on one shared component, so they look and behave identically and improve together. (A role granted attachment access without attachment-note access will now see note actions disabled.)

24 June 2026 — Comply AI on intake forms, and an optional address group

What's new

• Comply AI on intake-form pages. The Comply AI assistant is now available on intake-form pages — both a form's detail page and the form editor — so you can ask about a form while building or reviewing it, the same way you already can on profiles, alerts, and blockchain addresses.

What's fixed

• Choosing a group is optional. You can save a blockchain address without a group, matching the rest of the platform, and clear a previously assigned group back to none from the address Overview.

16 June 2026 — Blockchain-address groups, and "Profile Groups" becomes "Groups"

What's new

• Addresses belong to a group. Blockchain addresses now belong to a governing Group, the same way profiles do. When you add an address on its own you choose its group (auto-selected when your org has only one), and that group drives the address's risk thresholds and screening schedule; addresses created within a profile inherit that profile's group. The group shows as an editable, linked row in the address Overview and follows the usual edit-and-save and rationale rules.
• Deep links switch organization for you. Opening a link or bookmark to a profile, blockchain address, alert, intake form, or group that lives in a different organization now switches your active organization automatically to open it, instead of failing to load.

What's improved

• "Profile Groups" is now simply "Groups." The feature is renamed across the app — every "Profile Group" label, and the settings section, now reads "Groups," and its settings page moved from /settings/profile-groups to /settings/groups (old links redirect automatically). Now that both profiles and blockchain addresses belong to one, the shorter name fits what the feature actually governs.

What to watch

• For API and AI-tool users, this is a breaking rename. The same change renames the API and MCP surface: the .../profile-groups/ endpoint becomes .../groups/; fields rename (profile_group → group, profile_group_name → group_name, profile_groups_url → groups_url); the ?profile_group__in= / analytics ?profile_group= filters become ?group__in= / ?group=; profile_screening_min_score becomes screening_min_score; the profile_group.* permissions become group.*; and the MCP tools profile_group_* become group_*. Anything hard-coding the old paths, fields, permissions, or tool names must update. (The intake-form survey key ca_profile_group is intentionally unchanged.)

12 June 2026 — Dropdowns open in front of dialogs

What's fixed

• Selection dropdowns always appear in front. The risk-threshold editor and the risk-engine group and blockchain-address dialogs previously opened their dropdowns behind the dialog, leaving options invisible and unclickable. They now always open in front.

11 June 2026 — Branded error pages, and long notes no longer vanish

What's new

• Branded error pages. When a page can't be found or something goes wrong — a missing profile or address, a broken link, or a server error — you now see a clear, branded in-app screen (including a dedicated "not found" page for a record that doesn't exist) instead of a blank or raw browser error.

What's fixed

• Pasting a long note no longer saves an empty one. The character counter now always reflects the editor; when the text exceeds the limit it turns red and disables save until trimmed. The limit was doubled to 10,000 characters so longer compliance notes fit.

9 June 2026 — Live change highlights and custom profile fields

What's new

• Live change highlights. When a teammate, an AI assistant, or an automated screening updates a profile or address, the exact fields, sections, tabs, badges, and rows that changed briefly flash so you can see what moved. Highlights are scoped to the record you're viewing and never flash your own edits.
• Custom profile fields. Custom fields configured for your organization now appear in a profile's Primary Party Information alongside the standard fields — shown in read mode, editable inline, and captured automatically from matching intake-form questions.
• Submission confirmation page. After submitting any intake form (public or invitation-only), the submitter sees a branded "received securely" confirmation instead of the form quietly resetting.

What's fixed

• Blockchain picker shows networks again — the "Select blockchain" dropdown opens in front of the dialog and table filter panel, so the list of networks is visible when adding an address, in profile creation, and when filtering the address table.

4 June 2026 — Stay signed in, notification types, and alert rationale

What's new

• New notification types. The notification center now recognizes alert-closed notifications, intake-form profile-onboarding notifications, and split private/public intake-invitation events.

What's improved

• Alert assignment rationale. Assigning an alert to another reviewer now captures a short rationale, delivered with their notification and shared with the take-over flow. Self-assignment is unchanged.
• Live refresh after profile edits. Editing a profile automatically refreshes its screening status and compliance health, without a manual reload.
• Streamlined profile Actions menu — the redundant "Edit Profile" and "Screening hits" shortcuts (which just jumped to views already on the page) were removed.

What's fixed

• Stay signed in. The app now refreshes an expired session in the background instead of signing you out after a few minutes of normal use.
• KYC document links open the file from an alert's side panel, instead of an internal API page.
• Duplicate wallet address flagged inline on the address field with a clear message, rather than only as a passing error.

30 May 2026 — Request for Information, redesigned

What's new

• Redesigned RFI workflow. "Send RFI" and "View RFIs" live on the profile Actions menu, gated by dedicated RFI permissions. The request form is schema-driven, and a history drawer shows every RFI sent for a profile along with what was requested.
• Response review and adjudication. When a client responds, a review banner appears on the profile; reviewers see each submitted answer next to its current value and accept or reject it item by item — nothing changes on the profile until they do.
• Email preview. The send modal can preview the exact email before it goes out, and the history drawer can reopen the email that was already sent.

What's fixed

• Periodic code security audit completed.

28 May 2026 — Redesigned alerts workflow

What's new

• Alert detail page. Each alert opens to a dedicated detail page sized for the decision: watchlist alerts get a hit-by-hit comparison panel; blockchain-address alerts get a risk-change breakdown with new exposures highlighted. A side panel collects the alert's audit trail, decision form, and related alerts, and the lifecycle actions — Claim, Assign, Unassign, Close (with disposition), Escalate (with recipient + reason) — are grouped on the alert header.
• Profile Alerts tab. The profile page has a new Alerts tab listing every alert linked to the profile, plus a banner summarizing open alerts for related parties (UBOs, controllers, officers) so an EDD reviewer can pivot laterally without leaving the profile.

What's improved

• Alerts list with inline filtering. The alerts list surfaces risk tier, sanctions/PEP/adverse-media flags, and watchlist source name as filterable columns.
• Take-over requires a rationale. Taking over an alert assigned to another analyst now requires a written justification, recorded on the audit log.

26 May 2026 — Reliable loads and quicker tab counters

A smaller follow-up: a fix so a freshly published release is always picked up on the next visit, plus faster tab counters on profile pages.

What's new

• Notes and attachments counts up front. Profile pages now show how many notes and attachments are attached on the tab labels without first downloading the full lists, so the counters appear immediately.

What's fixed

• No more blank pages after a release. After a new release goes out, opening a deep link such as /login or a specific profile page used to occasionally serve a stale, blank screen for a few minutes. The latest release is now picked up immediately on the next visit.
• Profile status changes from the overview save correctly. Editing the Review Status or Relationship Status from the profile overview now persists — previously the success message appeared but the value silently reverted on refresh.

22 May 2026 — Cross-profile analysis via your AI assistant

A major expansion of what your AI assistant — Comply AI in-app, or any external tool like Claude or ChatGPT connected via MCP — can do across your organization's records, plus a couple of UI-visible improvements to profile creation and archiving.

The capabilities listed under What you can ask your AI assistant are not buttons in the app. They become available when you talk to Comply AI in-app, or to any MCP-connected AI tool such as Claude Desktop or ChatGPT. See the MCP integration guide.

What you can ask your AI assistant

• Bulk profile review at a glance. Ask your AI assistant to pull a full table of profiles in your organization — with wallet, watchlist, adverse-media, and RFI data joined inline — in a single request. Reviewing hundreds of profiles is dramatically faster than reconciling four separate views by hand.
• Stuck-RFI detection. Ask your AI assistant to flag profiles that are currently in RFI Sent status but have no RFI actually dispatched since they entered that status — catching the case where status was flipped manually without ever sending the request.
• Duplicate detection. Ask your AI assistant to surface profiles that share a phone number, an email domain, or a bio URL (with normalization), grouping them as clusters that may indicate coordinated registrations or proxy accounts.
• Identity-signal detection. Ask your AI assistant to flag credential anomalies — an email handle that looks like a phone number, a generic English first-name handle on a company domain, a handle that appears in another profile's name — as leads to investigate, each with a low / medium / high confidence indicator.
• Bulk actions at scale. Through your AI assistant, move up to 100 profiles to a new review status in a single step (the system walks each profile through the legal path of the status workflow), or add the same note to up to 500 profiles at once.
• Tag operations. Your AI assistant can now add, remove, or replace tags on profiles — strip every existing tag and install a new set, or remove a specific subset, in one call.
• Analytics drill-down. When you ask your AI assistant for an analytics summary, it can now break results down by profile group or risk level in the same answer.

What's new in the app

• Provision a profile in one step. A profile can be fully set up at the moment you create it — schedule, screening configuration, tags, and rationale included — with no follow-up edit required.
• Archive from any status. A profile can now be archived from any review status, instead of needing to be walked through intermediate transitions first.

What's improved

• Unified screening results. Watchlist, sanctions, and PEP results from every screening provider now look the same on screen, so a hit from one provider reads the same as a hit from another.
• Faster cold loads. The first page you open after a period of inactivity now loads noticeably faster.
• Faster analytics dashboards. Organization analytics now load faster, especially when several teammates are viewing the same dashboard at the same time.
• Polished notification emails. Notification emails refreshed for clarity, and admins can now configure many notification events in a single step instead of one at a time.

What's fixed

• No more false Profile Approved notifications during bulk updates. If a bulk status change ran into a problem mid-way and rolled back, you could still receive notifications for the intermediate steps that didn't actually stick. Notifications now only fire after the change has fully completed.
• Invalid status transitions are blocked everywhere. A few less-common ways of updating a profile (through an API key or an AI tool) could previously skip the validation that prevents illegal status changes (for example, jumping straight from Pending Review to Approved without a review). Those paths now apply the same validation as the web app.
• Faster loading on profiles with many notes. Notes lists on profiles with hundreds of notes now load quickly, instead of trying to fetch the full set at once.
• Threshold-update notifications no longer slow down on large organizations. When a risk threshold changes in an organization with many members, sending out the resulting notifications is now noticeably faster.

15 May 2026 — Live updates everywhere

Profiles, alerts, blockchain addresses, notes, and tags now refresh in place whenever something changes — without a page reload.

What's new

• Live updates across the app. Edits land instantly across every connected session: when a teammate makes a change in another tab, when an MCP client (Claude Desktop, Comply AI, etc.) updates a record, when an automated screening completes, or when an external integration posts an update, the affected views re-fetch the moment the change lands.
• No unnecessary re-fetches on your own edits. Your own tab does not refresh again on changes you just made — only changes from other people, AI tools, or integrations trigger a refresh in your view.
• Rationale enforcement everywhere. Audit-rationale requirements are now applied the same way regardless of channel — the web app, AI tools, and external integrations all run the same check. The rule also applies to Review Status and Relationship Status transitions, not only field edits.

What's improved

• Audit log attributes API-key activity to the right person. Changes made through a per-user API key now record the key's owner as the actor in audit logs, instead of attributing them to the system. Audit dashboards filtering for system-actor entries to find API-key activity may need to be reviewed.

14 May 2026 — Comply AI assistant and external AI tools

The launch of the in-app AI assistant, a way to connect outside AI tools, and a major sign-in refresh.

What's new

• Comply AI assistant. Ask compliance questions in plain language — "which high-risk profiles do we have?", "show me open alerts assigned to me", "summarize this entity's blockchain exposure" — and get answers drawn from real CryptoComply records, never guessed. Responses stream back in real time, with each step the assistant takes (the tools it called, the records it consulted) shown inline in a collapsible track so analysts can audit the reasoning, not just the answer.
• Conversations saved per user. Conversations are saved per user per organization, automatically titled, and searchable. Return to any past session from the sidebar, rename it, or pick up where you left off.
• Edit and regenerate. Any previous message in a conversation can be edited; the assistant regenerates its response from that point without starting a new conversation.
• Cancel in-progress responses. A cancel button appears while the assistant is generating, so a long response can be interrupted at any time.
• Feedback on every answer. Thumbs-up and thumbs-down on individual responses help improve answer quality.
• Markdown formatting. Assistant responses render full markdown — headers, lists, bold, code blocks, and tables — instead of plain text.
• Connect external AI tools. Tools like Claude Desktop, Claude Code, claude.ai, ChatGPT, and Goose can now connect securely to CryptoComply over the internet. Each tool signs in as a real user through the browser, inherits that user's permissions and organization memberships, and operates on real records.
• Write actions require explicit confirmation. When an external AI tool tries to update a profile, add a note, resolve an alert, or change anything else, the action is shown for review before it touches the database. Approval is required every time; AI tools cannot silently mutate data.
• Profile, alert, address, RFI, screening, and tag operations from AI tools. External AI clients can browse, filter, view, and update profiles (with status, screening progress, and risk history available), browse and resolve alerts, manage blockchain addresses (with full exposure breakdowns), create RFIs, trigger screenings, and create or assign tags.
• Notification subscription audit. Org admins can see from Settings → Notifications who in their organization receives each notification type and through which channels, in two views: by event (every notification with the subscribers who will receive it) and by user (any member's full notification matrix at a glance).

What's improved

• Sign-in screen shows the active account. The sign-in screen now displays the logged-in user's email prominently, with a Not you? Switch user link to sign in as a different account without leaving the flow.

What's fixed

• Google sign-in restored. Sign-in with Google was being hidden by a default template; the button now works correctly on both the login and authorization pages.

Heads up

• /dashboard prefix dropped. Web addresses no longer carry the /dashboard prefix — old bookmarks continue to work via automatic redirect.
• Previous chat assistant replaced. The earlier AI chat (which routed messages through an external workflow service) has been retired in favor of the new in-app assistant. Existing conversations were migrated automatically — no user action required.
• Some personal notifications no longer configurable. Alert assigned to you and the bulk-invitation completion notice no longer appear in notification settings — they always fire to the targeted user.

---

Looking for older releases? See the What's New archive — April 2026 and earlier.