AckuityRequest assessment
Menu

ROGUE PIXEL · Reported incident

One login opened more doors.

The final control-switch sequence is proposed, not an observed intervention. GitBook attributes the attack to automation and AI; its own AI products are not identified as the cause.

Read the narration

GitBook says AI agents hit its docs platform. An SSO flaw made staff sessions, a Cloudflare key, then production secrets. Offline on purpose. Every API token invalidated. The check sits outside the agent and judges each action before it runs.

Read the visual story

A generic documentation platform shows an access cascade through staff sessions, repositories, a Cloudflare key and production secrets. Token invalidation and an outage follow. The final independent action check illustrates a proposed safeguard.

By Ackuity · Published · Event: 2026-11-02

GitBook says an SSO flaw led to staff sessions, code repositories and production secrets. It deliberately took the app offline and invalidated existing API tokens.

THE CONTROL LESSON

What should we watch for?

Ackuity’s view: check actions against the authorized task, identity, credentials and target before execution. Keep enforcement outside the acting agent’s control.

THE EVIDENCE

What does the report establish?

AI involvement is GitBook’s assessment of the attack’s speed and precision; no attacker model is named. GitBook treated all customer data as exposed, while reporting no evidence of other customers’ content being modified. Accessible data is not proof that every item was taken. The animation’s independent check is proposed, not an Ackuity intervention.

Read the full account

GitBook’s November 4 postmortem traces the attack from its SSO flow through a misconfigured GitHub App and a Cloudflare API key. Compromised service accounts were disabled; secrets were rotated. Service returned after about 16 hours.

A credential can connect several systems. Each new action needs authorization appropriate to its target.

ORIGINAL REPORTING & EVIDENCE

PUT THE LESSON TO WORK

What can one compromised credential authorize next?

Trace the systems reachable from each identity and integration key. Test whether an independent check can reject an action outside the approved task, and verify the target-system outcome alongside the decision record.

Test credential scope and action evidence

Explore Ackuity’s Agent Execution Control Switch

Request an Agent Exposure Assessment