GitBook agentic AI attack — Nov 2024

GitBook, a documentation platform frequently utilized by cryptocurrency projects, disclosed a security breach on November 2, 2024, involving the use of AI agents. The incident resulted in unauthorized access to infrastructure credentials, prompting the company to take services offline and mandate that customers rotate their digital keys.

What happened

According to GitBook, the intrusion occurred at 7 p.m. Central European Time on November 2. The company characterized the event as a sophisticated, highly targeted attack that leveraged AI agents to gain unauthorized access. Unlike standard chatbots, these agents were capable of performing sequences of actions and interacting with software with limited human supervision.

GitBook reported that the attacker gained access to several categories of sensitive information, including API tokens, integration secrets, and questions submitted to the platform's AI assistant. The company stated it has found no evidence that hosted content was modified during the breach. To contain the incident, GitBook took its application and documentation sites offline, invalidated personal API tokens, and replaced its own infrastructure credentials.

While GitBook has not disclosed the specific AI systems involved or the exact method used to obtain the credentials, it emphasized that the breach did not necessarily involve its own AI products. The incident impacted various integrations, including Git Sync secrets used for connecting documentation to code repositories and credentials for the Model Context Protocol (MCP) server.

In response, the cryptocurrency exchange PancakeSwap temporarily took its documentation offline as a precaution, later migrating its documentation to infrastructure under its own control. By Tuesday afternoon, GitBook reported that its application, API, and documentation sites were accessible again, though it urged all users to rotate existing API tokens and replace third-party secrets connected to the platform.