StackJack 0.1.0.13653+c3fa33a
StackJack 0.1.0.13653+c3fa33a
This release focuses on AI Governance exposure scans. Scans now finish on large organizations, find AI apps more reliably, and produce clearer, more accurate reports. It also fixes Microsoft Copilot source reconnection and team invitations.
New features
- AI app detection no longer depends on a fixed list. Exposure scans now use an AI-based assessment to decide whether each outside app in a directory is AI-related. Newly released AI apps are recognized without waiting for a list update. Name-based matching still runs as a second signal, and either signal can mark an app as AI-related. If the AI assessment is unavailable, the scan continues with name-based matching, and the report says so in one line.
- New finding for outside apps with broad access. A non-Microsoft app that is not identified as AI-related but holds a high-impact permission or role now raises a Medium finding. The finding notes that the scan cannot tell what the app does.
- Per-surface Copilot activity counts. Reports now break down prompt injection attempts, jailbreak attempts, labeled resources and the number of distinct people for each Copilot surface. Scans stored before this release show their totals without the per-surface table.
Improvements
- Exposure scans complete on large organizations. Previously, scans on large directories could stop partway through sign-ins, audit records and detected device apps. Time limits and read limits have been raised substantially. Scans now wait and retry when Microsoft asks them to slow down, instead of failing that part of the scan.
- Scans survive platform updates. A scan that is running during a StackJack update goes back into the queue and runs again, instead of failing or saving incomplete results.
- Comparisons use a valid baseline. A baseline scan taken before the earlier exposure scan update could not be compared against, so later comparisons had nothing to compare to. The next complete scan now becomes the baseline automatically. The older scan keeps all of its results.
- More accurate privacy statement. The report now states exactly what is sent for the AI assessment: each app's name and related app details. It also states that no person's name, file or message is sent, and it gives the number of apps that were sent. "The AI check did not run" now appears only when no apps were actually sent.
- StackJack's own apps are excluded. StackJack's own connector and directory sync apps no longer appear in app lists or counts.
- Copilot sources return after a Microsoft reconnect. When you disconnect and reconnect to the same directory, Copilot sources that the disconnect switched off are switched back on. Sources you turned off yourself stay off. Adding a source again reuses the existing one instead of creating a duplicate, and existing duplicates are cleaned up from the list.
- Faster staff list updates. Changes to your own staff list are now picked up within about 30 minutes, instead of up to 6 hours.
- Scans stored before this release keep the risk ratings they were given. The new detection rules apply to new scans.
Bug fixes
- Fixed exposure scans failing to read the Global Administrators list.
- Fixed the evidence table under a finding showing unrelated data. Each finding now shows only the data that supports it, in the Portal, the presentation view, the PDF and assistant answers.
- Fixed counts from partially read data appearing as exact numbers. These counts now appear as minimums, and a zero from incomplete data shows as "Not fully read". This applies to usage, desktop app, agent and outside-app tables.
- Fixed the agent table showing agents found only in audit logs as "Turned on".
- Fixed the agent table adding up people across several agents. The combined figure is now shown as a minimum, because the same person can appear under more than one agent.
- Fixed invitations to an email address whose domain is claimed by another organization failing with "could not be created" on invite and on resend. The invitation now gives an accurate result.
- Fixed report and evidence links not opening reliably. They now download the file directly.
- Fixed a long unbroken piece of text making the stored text dialog scroll sideways indefinitely.
- Fixed a reconnect in a client organization that is not active switching its sources back on. Those sources now stay off.
UI & UX changes
- When more than three parts of a scan could not be compared, the report now shows one summary sentence instead of a separate line for each part.
- The Microsoft page now shows an On or Off control, instead of Add, for sources that already exist.
- A source that has been replaced after a reconnect is hidden from running-state pages. It still appears, marked as replaced, in the Activity and Findings pickers so you can filter older records.