GitBook hit by AI-agent cyberattack targeting infrastructure

GitBook disclosed a targeted cyberattack involving AI agents that gained unauthorized access to infrastructure credentials on November 2, 2026 GitBook. The documentation platform said the attack was sophisticated and highly targeted, exploiting a flaw in its SSO login flow to create sessions acting as GitBook staff members.

How the attack unfolded

The attacker leveraged the SSO vulnerability to gain unauthorized access at approximately 19:00 CET, then used a stolen credential to access production secrets through a service account Yahoo Cybersecurity. This move granted the attacker full infrastructure access, exposing GitBook's core systems and sensitive data stores.

What was exposed

GitBook identified several categories of potentially compromised information, including integration tokens, Git Sync secrets, GitBook API tokens, third-party API tokens, and questions submitted to GitBook Assistant Yahoo Cybersecurity. The breadth of exposed credentials created cascading risk across GitBook's customer ecosystem, since many users integrate GitBook documentation with external services.

Containment and customer response

GitBook took its application and authenticated documentation sites offline while investigating and securing its systems. The company revoked all personal and customer API tokens, then instructed users to rotate GitBook API tokens, third-party tokens, integration secrets, and other connected credentials GitBook. Users of GitBook's Model Context Protocol server were separately instructed to re-authenticate.

The attack marks a notable incident in the emerging category of AI-agent-driven cybercrimes. Rather than relying solely on human attackers or traditional malware, the breach leveraged AI agents to automate reconnaissance, credential exploitation, and lateral movement through GitBook's infrastructure—a pattern that security researchers have begun tracking as offensive AI capabilities mature.

GitBook said it blocked the attacker, closed the underlying vulnerabilities, and replaced all infrastructure credentials and secrets Times of AI. The company restored services after containment was complete but did not disclose a specific timeline for full restoration or provide an assessment of whether customer data was exfiltrated beyond the credential types already identified.