Red Teaming Changelog

What changed in Red Teaming, newest first.

Detect AI workflows from a URL (2026-10-30)

On a new AI Workflow assessment, paste the workflow's upload URL or host and click Detect, or choose Advanced / configure manually to fill in every setting yourself
Detect reads the workflow's API description and uploads a harmless sample document to fill in the upload settings, including polling for workflows that run as jobs
Check the filled-in settings, add the credential if the workflow needs one, then test and launch

The Recent runs list on the Red Teaming Assessments page shows how many runs are running right now
Click the count to show only the running runs

Forward summary on a run's page sends the pass rate, findings by severity and the top findings to a Slack channel, to email addresses, or both
Each message links to the full results in the console, and people sign in to see them
Nothing is saved: paste a Slack incoming-webhook URL and up to 10 email addresses each time

Paste a web page or a bare host and Detect follows the host's own leads to the agent's API: A2A agent cards, OpenAI plugin manifests, OpenAPI and Swagger documents, OpenAI-compatible /v1/models listings, the scripts on the page, and common chat paths
It fills in the request body and the reply field, and reads streamed (text/event-stream) replies
Other endpoints found lists the rest of what it found; Detect this one tries another endpoint on the same host
Detect reads how the API signs callers in (its sign-in challenge, OpenAPI security schemes, OAuth and OpenID Connect metadata) and pre-fills the most likely method, saying where it came from. Methods it can't use, such as NTLM, AWS request signing or SSO, show a warning instead
New sign-in methods: API key in a header or the query string, HTTP Basic, and a sign-in form that keeps its session cookie, with optional anti-CSRF token support
Discovery only reads from the host you pasted, with at most 30 requests in 45 seconds, and sends one test message to the endpoint it picks
Details: Connect your agent

Each framework card on the Standards & Framework Coverage panel now shows the pass rate of the run's tests under that framework: green when they passed, red when they failed. A test that maps to several controls counts once, and tests that errored are left out
A large Overall pass rate ring sits beside the framework cards
The panel header shows how many test cases passed out of the total, for example Tests passed 12/14
The line under each ring shows how much of the framework the run reached, for example 7 / 18 techniques tested. A Quick scan reaches less than the Full library
Framework cards use the same icon tiles as the New assessment cards
Details: Reading a report

The AI Workflow card on New assessment can now be started: test a document, image or structured-data workflow that processes inputs and returns a result or takes an action
Test connection checks the workflow endpoint before you launch