GitBook, a documentation platform widely used by cryptocurrency projects, has disclosed a security breach in which an attacker used AI agents to gain unauthorized access to some of its infrastructure credentials, forcing the company to take services offline and revoke customer access tokens.

In a security disclosure published on November 3, the company described the incident as "a sophisticated, highly targeted attack using AI agents." According to GitBook, the attacker gained unauthorized access at 7 p.m. Central European Time on November 2, and the intrusion was detected immediately and contained. The disclosure is available on GitBook's security incident status page, with related updates posted to its status site.

The reference to AI agents is significant. Unlike traditional chatbots that simply respond to questions, AI agents can carry out sequences of actions, interact with software, and perform tasks with limited human supervision, making them potentially useful to attackers seeking to automate parts of a cyberattack. Some are even capable of acting autonomously and against their guidelines, as experiments by OpenAI and Anthropic have demonstrated.

GitBook has not explained exactly how the attacker used these tools, which AI systems were involved, or how the credentials were obtained. Its disclosure also does not establish that GitBook's own AI products were responsible for the intrusion. Those unanswered details remain the key open questions as GitBook continues posting updates to its status pages.

While securing its infrastructure, the company took its application and documentation sites requiring authentication offline. Publicly available documentation remained accessible through cached copies for as long as possible, allowing visitors to continue reading previously saved versions of pages even while the underlying systems were unavailable. A separate status notice acknowledged downtime affecting published documentation sites and the editor. By Tuesday afternoon, GitBook said its application, API, and published documentation were accessible again, although some services could remain degraded during recovery.

Potentially exposed information includes API tokens, integration secrets, and questions submitted to GitBook's AI assistant. The company said it has found no evidence that hosted content was modified.

What Was Exposed, and Why It Matters

GitBook effectively functions as an online instruction manual for software products. Cryptocurrency projects use platforms like it to publish technical explanations, developer instructions, and guides covering everything from connecting software to interacting with decentralized financial applications. The documents themselves are often public, but the systems that maintain them can connect to private accounts and external services. Those connections rely on credentials that are supposed to remain secret, and GitBook says some of them may have been exposed.

Its investigation identified several potentially affected categories, including integration tokens, Git Sync secrets used to connect documentation with code repositories, and API tokens that allow software to communicate with GitBook automatically.

Crypto Projects Take Precautions

The incident has already prompted a response from the cryptocurrency industry, where documentation websites often serve as an important source of official instructions. Decentralized exchange PancakeSwap temporarily disabled its documentation following news of the breach, although it said its own systems had not been compromised.

In a public statement, PancakeSwap said its "docs are not compromised and [the] PancakeSwap app is not affected." The exchange nevertheless took its documentation offline as a precaution and redirected visitors to its homepage. PancakeSwap also reminded users that its documentation does not require them to perform wallet transactions, warning that "docs never ask you to connect your wallet or approve tokens."

PancakeSwap subsequently announced that its documentation was back online after being moved from GitBook to infrastructure under its own control, with existing documentation links continuing to work. The episode shows how an incident at a shared vendor can lead even unaffected projects to rethink where their public documentation is hosted.

GitBook Tells Users to Replace Their Digital Keys

GitBook said it blocked the attacker, addressed the vulnerabilities involved, replaced its own infrastructure credentials, and invalidated customer API tokens. Customers who use GitBook alongside other services, however, still have work to do. Those third-party credentials are configured by customers themselves, sit outside GitBook's control, and can only be replaced by whoever created them.

The company's instructions begin with "rotate any existing GitBook API tokens," meaning customers should replace their old digital access keys with newly generated ones. Rotation matters because issuing a new token invalidates the old one, cutting off any further use of a credential that may have been exposed. GitBook also instructed customers to replace third-party API tokens and secrets used to connect external tools to its platform, including credentials associated with published integrations. Users of its Model Context Protocol (MCP) server must authenticate again; MCP is a system that allows AI assistants and other software tools to connect to services such as GitBook.