In previous versions it was possible to create an attack campaign in the UI with no attacks. Now users are prevented from saving a campaign when no attack vectors or prompt packs are selected.

In some cases, filtering by tags on the Scanners page returned incorrect results. This has been fixed.

In some situations, Chat users could see an “Undefined” model instead of the model they had chosen earlier. This issue has been corrected.

Users with a custom role where all permissions were enabled encountered error messages on various pages. This issues has been fixed.

We’ve fixed the scanner card for customer regex scanners so that the description area shows the actual regex code.

There was an issue in the Raw Data View for the Red-Team report where the prompt and response columns for operational attacks were empty, even though the data appeared in the download. This has been fixed.

We corrected an that caused some Agentic Warfare custom campaigns to be incorrectly saved as the “Standard” type.

Some users reported seeing duplicate versions of the same scanners. This issue has been tracked down and fixed.has been tracked down and fixed.

In Signature attack campaigns, it was possible to disable all three prompt packs, however the system would still run the latest. Now the UI clearly messages that the latest prompt pack will be run.

We’ve changed the default order of attack results in the Raw Data Table and report download so that Signature attacks are listed first, and Operational attacks last.

In Chat, when choosing a project with no providers, users could not change their project selection. Now they can.

The filters and sorting on the Custom Scanners table were resetting after a user performed an action, such as deleting a scanner. The expected behavior in this case is that filters and sorting persist; this has been fixed.

We fixed a filtering issue in the Playground UI that caused individual scanners to be shown when the user had attempted to filter by packaged scanners only.

User-created scanner tags were not being displayed in the Playground. Now they are.

There was an issue where the CASI score was not showing in the Reports table, though the score was visible on the Report detail. This has been fixed.

We’ve added a loading indicator to the Project detail so there is feedback to the user when the data is still loading.

We fixed a UI bug so that converters are unchecked when the parent attack vector is unchecked

In some reports the Mitigation Actions count was incorrect. We’ve identified the source of the miscount and correct the problem.

API users now have the ability to delete campaign runs.

In the Red-Team attack scheduler, users in different timezones were encountering a bug that made it difficult to select the desired time. This has been fixed.

In the Red-Team report XLS download, the prompt and response columns for operational attacks were empty. They are now populated with the correct data.

After deleting all scanners in the Playground, it was possible to still see them all by using the Filters function. This has been fixed so that deleted scanners stay deleted.

We resolved a UI bug that caused a misleading “Enforce” checked state when a scanner or scanner package was inactive.

Users who tried to connect to an image model were seeing errors in the Red-Team report. This is because image models are not supported on the platform as yet. To avoid confusion, we’ve restricted Connections to text models exclusively.

Chat Projects were missing a tooltip to identify their icon. We’ve added the tooltip.

We made UI improvements to correct colors and element placement on the Chat and Custom Roles screens, and fixed a funky click area in the Create Project wizard.

In earlier versions, the All Attacks campaign was incorrectly showing all available prompt packs as selected. This has been fixed so that only the latest prompt pack is selected by default.

Scanners can only be unpublished one at a time, even though the UI suggests it’s possible to bulk unpublish by selecting multiple scanner checkboxes.

In the Create Projects wizard, model provider icons are the wrong size when the model name is long.

The scanner counts on Project tiles can be incorrect in some cases.

The “Load more” button on the Playground doesn’t take applied filters into account.

In the unusual case where a CalypsoAI scanner package has global access turned off AND is not assigned to any Project, that package will appear in the Project to be available even though it isn’t.