Skip to content

Back to the catalog

Crypto Docs Host GitBook Hit by AI Agent Security Breach, Urges Users to Rotate Keys
decrypt.coarticle

Crypto Docs Host GitBook Hit by AI Agent Security Breach, Urges Users to Rotate Keys

GitBook disclosed an attack it calls "sophisticated, highly targeted" and AI-agent-assisted. The attacker gained unauthorized access to infrastructure credentials at 7 p.m. CET on 2 November 2026. Potentially exposed data includes API tokens, integration secrets, Git Sync secrets and AI assistant questions. GitBook revoked personal API tokens and told customers to rotate credentials and re-authenticate MCP. PancakeSwap moved its docs off GitBook. GitBook has not said how the agents were used.

TL;DR Reports GitBook's 2 November 2026 breach, said to involve AI agents, that exposed API and integration tokens and forced customers to rotate keys.
Document

What it is

GitBook, a documentation host widely used by crypto projects, disclosed on 3 November 2026 an attack it calls "a sophisticated, highly targeted attack using AI agents". The attacker gained unauthorized access at 7 p.m. Central European Time on 2 November 2026. GitBook says it detected the intrusion immediately and contained it.

Preconditions / Affected

Anyone with a GitBook account that holds credentials GitBook may have stored. GitBook lists these potentially exposed categories:

  • API tokens that let software talk to GitBook.
  • Integration tokens and secrets.
  • Git Sync secrets, which link documentation to code repositories.
  • Questions submitted to GitBook's AI assistant.

Published docs are often public, but the credentials connecting them to private accounts and external services are not.

What to do

GitBook's instructions, in its stated order:

  1. Rotate any existing GitBook API tokens. GitBook has already invalidated customer API tokens.
  2. Replace third-party API tokens and secrets used to connect external tools to GitBook, including credentials tied to published integrations.
  3. Re-authenticate if you use the GitBook Model Context Protocol (MCP) server.

The page does not name Git Sync secrets in the action list. Since GitBook lists them as potentially exposed, rotating them on the repository side is a reasonable reading, but that is an inference.

Caveats

  • GitBook has not said how the agents were used, which AI systems were involved, or how the credentials were obtained.
  • The disclosure does not establish that GitBook's own AI products caused the intrusion.
  • GitBook says it found no evidence that hosted content was modified.
  • GitBook says it blocked the attacker, fixed the vulnerabilities involved and replaced its own infrastructure credentials. The page does not describe those vulnerabilities.
  • Authenticated app and docs sites went offline during containment. Public docs were served from cached copies where possible. By Tuesday afternoon, GitBook said the app, API and published docs were reachable again, though some services could stay degraded.
  • PancakeSwap took its docs offline as a precaution and said its docs and app were not compromised. It then moved the docs to infrastructure it controls, and existing links still work.

Organisations named here

  • Anthropic
  • OpenAI
  • PancakeSwap

Technologies named here

  • Model Context Protocol server
  • Model Context Protocol
  • GitBook
  • GitBook integrations and Git Sync
  • Git Sync
  • PancakeSwap documentation (hosted on GitBook, temporarily taken offline as a precaution, then moved to PancakeSwap's own infrastructure; PancakeSwap says its docs and app were not compromised)

Techniques named here

  • Credential theft
  • AI agent-assisted cyberattack

Concepts named here

  • AI agents
  • Credential rotation
  • Documentation platform supply dependency