AI-Driven Attacker Turns One SSO Flaw Into Full Access at GitBook, 16-Hour Outage Follows

On November 2, 2026, an attacker broke into GitBook, the documentation platform many software companies use to publish their help pages and developer docs. GitBook calls it "a sophisticated, highly targeted attack using AI agents."

According to GitBook's postmortem, as reported by TokenPost, the attacker exploited a flaw in GitBook's single sign-on (SSO) login to forge an employee session. From there, credentials from a misconfigured GitHub App opened GitBook's internal code repositories, where the attacker found a Cloudflare API key, then used a service account to reach production. The attacker spent about three hours and 25 minutes inside production systems.

To cut the attacker off, GitBook took its platform offline for about 16 hours, replaced every credential and secret across its infrastructure, and revoked every customer's API tokens. Full service came back on November 3.

GitBook is treating all customer data as potentially exposed: API and integration tokens, Git Sync secrets, session-signing keys, the questions visitors asked GitBook Assistant, and GitBook's app permissions inside customers' GitHub and GitLab repositories. It says it found no evidence that customer content was changed. A security researcher at SlowMist named crypto trading platform Trade.xyz as the targeted customer; Trade.xyz says its own systems show no sign of compromise.