Release 2026.10.28-1
Processing warnings everywhere you look at a transaction, a smaller attack surface for page images and the runtime image, and the security advisories linked from Support.
- Cloud, United States
- Cloud, Australia
- Self-hosted image
- Not in this release
Security
- Page and thumbnail images are fetched with the session header only: a session token no longer appears in an image address or in a log. A browser tab left open across this update shows no page images until it is reloaded.
- The runtime image no longer ships a package manager.
- Names in invitation and organization emails are escaped.
- DNS library updated.
New
- Processing warnings: a transaction that finished with warnings is marked as such, counted on Monitoring and shown in everyone’s Activity Log; the Agent Review step and the Assistant read them.
- Support page: links to the security advisories and to Updates and Security Advisories. The Trust Center lists the security advisories.
Improved
- The Activity Log shows one row per extraction, with the policies it ran with.
- When a model refuses a page because of its images, Extract tries once with the text only.
- Saving a tenant model checks that nobody changed it in between.
- The Assistant shows a translated message when its model call fails, and the Trust Center’s Assistant rows follow your organization’s Assistant settings.
API
da models: manage your organization’s own models from the command line. See Models.