Crypto HeadlinesStartup Cathedral Hacked Anthropic Systems Using AI Agents
Cathedral, a startup founded by former employees of the D.O.G.E. agency, successfully breached Anthropic's internal systems in less than 12 hours. Researchers used AI agents to find vulnerabilities in the corporate infrastructure of the developer behind the popular Claude model.
During the incident, attackers gained access to confidential information, including internal correspondence, meeting recordings, employee salary data, and discussions on advanced artificial intelligence models.
The incident sparked serious concern at the highest levels of government. A White House representative noted that if such a breach can be carried out by creators of a new startup, then similar threats are posed by other states, such as China.
Anthropic confirmed the fact of limited access to internal data. However, developers emphasized that source code, model weights, and personal correspondence of Claude users remained secure.
It is worth noting that the research was conducted within the framework of Anthropic's official vulnerability disclosure program (bug bounty), indicating the use of legitimate security testing methods.
