GitBook breach: attacker using AI agents accessed infrastructure credentials, tokens and assistant questions may have been exposed
GitBook said on November 3, 2026 that a sophisticated, highly targeted attack using AI agents gave an attacker access to some of its infrastructure credentials on November 2; integration tokens, Git Sync secrets, API tokens and questions visitors sent to GitBook Assistant may have been accessed, the company found no evidence hosted content was modified, and it revoked all personal API tokens and told customers to rotate their credentials.
What happened
GitBook, the documentation platform used by many software and cryptocurrency projects, disclosed on November 3, 2026 what it called “a sophisticated, highly targeted attack using AI agents.” The attacker gained unauthorised access to some of GitBook’s infrastructure credentials at 7 p.m. Central European Time on November 2, and GitBook says it detected the intrusion immediately and moved to contain it. The company took its app and authenticated documentation sites offline while it secured its infrastructure, kept public documentation available from cached copies where it could, and said on November 3 that its app, API and published documentation were accessible again, though some services could remain degraded. GitBook has not explained how the credentials were obtained, which AI systems were involved, or who was behind the attack.
Who is affected
GitBook customers, and visitors to documentation sites hosted on GitBook. GitBook says the attacker may have accessed integration tokens, Git Sync secrets (the credentials that connect documentation to code repositories), API tokens, and questions visitors typed into GitBook Assistant, the AI chat on customers’ documentation sites. The company says it found no evidence that hosted content was modified. It has given no count of affected customers or records. The cryptocurrency exchange PancakeSwap said its own systems were not compromised, but took its documentation offline as a precaution and later moved it to its own infrastructure.
How serious
The concrete risk is follow-on access through stolen credentials: an integration token, Git Sync secret or API token that still works can be used to reach the services it connects to — including code repositories — until it is replaced. Questions typed into GitBook Assistant are a second exposure: people sometimes paste internal details, links or credentials into a support chat, and anything of that kind in the questions may have been visible to the attacker. GitBook has not confirmed that any of this data was actually taken; the exposure is potential, and the company’s response — revoking every personal API token — is built for the worst case.
What to do now
Follow GitBook’s instructions: create fresh GitBook API tokens (the company has already revoked all personal tokens, so old ones no longer work). Replace third-party API tokens and secrets you connected to GitBook, and rotate the secrets used by any published integrations. If you use GitBook’s MCP server, re-authenticate it. If you typed anything sensitive into a GitBook Assistant chat on a documentation site — a password, key or internal link — treat that item as exposed and replace it where it is used.
Sources: Decrypt, reporting GitBook’s security disclosure, November 3, 2026 · The TokenPost, November 4, 2026