GitBook Reports AI-Agent Breach, Tells Users to Rotate API Tokens

GitBook, a documentation platform widely used by crypto projects to publish developer guides and technical docs, says an attacker used AI agents to gain unauthorized access to some of its infrastructure credentials, forcing it to take services offline and invalidate customer API tokens. The disclosure, published November 3, was reported by Decrypt, which cited GitBook's own security notice.

According to that disclosure, the intrusion began at 7 p.m. Central European Time on November 2. GitBook described the incident as "a sophisticated, highly targeted attack using AI agents," and says it detected and moved to contain the access immediately. The company has not explained how the attacker used those tools, which AI systems were involved, or how the credentials were obtained — and, by its own account, has not established that its own AI products were the way in.

Potentially exposed data includes API tokens, integration secrets, Git Sync secrets (used to connect docs to code repositories), and questions users submitted to GitBook's AI assistant, per the disclosure. GitBook says it found no evidence that hosted content was modified. It took authenticated application and documentation sites offline while securing infrastructure, kept public docs reachable via cached copies, and says that by Tuesday afternoon its application, API, and published docs were accessible again, though some services could remain degraded.

GitBook's remediation instructions, as relayed by Decrypt, start with "rotate any existing GitBook API tokens." It also tells customers to replace third-party tokens and secrets connecting external tools, including credentials tied to published integrations, and says users of its Model Context Protocol (MCP) server must re-authenticate.

Decentralized exchange PancakeSwap responded by temporarily taking its documentation offline, stating its "docs are not compromised and [the] PancakeSwap app is not affected," and reminding users that "docs never ask you to connect your wallet or approve tokens." It later said docs were back online after being moved off GitBook to its own infrastructure, with existing links still working.

Key facts

• Unauthorized access began 7 p.m. CET, November 2; disclosed November 3 (GitBook via Decrypt).
• Possibly exposed: API tokens, integration secrets, Git Sync secrets, AI-assistant questions; no evidence hosted content was altered (GitBook).
• Remediation: rotate GitBook API tokens, replace third-party tokens/secrets, re-authenticate MCP server (GitBook).
• PancakeSwap pulled docs as a precaution, says its systems were not compromised, then migrated docs off GitBook (PancakeSwap statement via Decrypt).

The real-world read

Much of this story rests on GitBook's own account, and the most important details are missing: the company has not said how the AI agents were used or how credentials leaked, and explicitly does not confirm its own AI products were involved. "Sophisticated, highly targeted" is the standard language of a breached vendor — treat it as framing, not fact, until technical specifics appear. The concrete, actionable part is unambiguous: if you connected anything to GitBook, assume those tokens are burned and rotate them. PancakeSwap's reminder that legitimate docs never ask you to connect a wallet is the useful takeaway, because incidents like this invite phishing that impersonates "updated" documentation.

This is reporting, not financial or security advice; verify your own exposure before acting.

Sources

• Decrypt (secondary, Nov 3) — reported the breach, the exposed data categories, GitBook's remediation steps, and PancakeSwap's response, citing GitBook's security disclosure and PancakeSwap's public statement. GitBook's disclosure is the primary account; its claims are self-reported and not independently verified here.