GitBook Hit by AI Agent Attack as Tokens Are Revoked and PancakeSwap Pulls Docs

Documentation hosting platform GitBook has been hit by a security attack involving an AI agent, according to reports from Coinfomania and Altcoin Buzz published on November 3. Coinfomania reported that GitBook implemented security measures in response, while Altcoin Buzz added two concrete follow-on actions in the same incident: tokens were revoked, and the crypto project PancakeSwap pulled documentation it hosted on the platform.

GitBook is a common home for API references, contract documentation and developer guides in crypto and open-source projects. When the documentation layer is compromised, the damage is not limited to a defaced page — developers who integrate against those docs lose a reference they had been told to trust.

The public record so far describes a sequence of actions rather than a full timeline: an attack, a token revocation, a project pulling its docs. Altcoin Buzz paired the revocations with PancakeSwap's removal in its headline, suggesting platform-side and project-side containment happened close together rather than as separate after-the-fact cleanups.

The important part of this story is the permission boundary around AI agents. When an agent is given read and write access to documentation repositories, admin APIs or the ability to act on a user's behalf, it holds not just a session but a set of credentials that can be used to move laterally. Revoking tokens is the fastest way to cut that access, which is why it appears to have been the first containment step.

Neither outlet's headline discloses the intrusion path, the number of affected accounts or the identity of the attackers, and the material reviewed here does not include a full incident report from GitBook. How an agent-related credential was obtained, and what exactly the agent could reach, remain unresolved questions.

What makes the case worth watching is that documentation platforms have traditionally been treated as a read-only presentation layer, with security budgets below those of code repositories and production systems. As agents take on editing docs, cutting releases and syncing API changes, those platforms get pulled into automated pipelines — and their credentials become worth stealing.

For teams running documentation on hosted platforms, the immediate lesson is to re-inventory what their agents hold: which credentials are long-lived, which can write rather than only read, and which can publish without human confirmation. Tightening scopes and rotation matters more than pulling pages after the fact.

Three things to watch next: whether GitBook publishes a complete post-mortem, when PancakeSwap restores its documentation, and whether other projects on GitBook disclose similar exposure.