Former employees of Musk's DOGE hack Anthropic's systems using AI

The Washington Post revealed that a startup founded by former employees of the U.S. Government Efficiency Office led by Elon Musk managed to hack into the systems of AI company Anthropic in less than 12 hours using AI agents, accessing sensitive internal data.

The Washington Post revealed that a cybersecurity startup, founded by former employees of the U.S. Government Efficiency Office led by Elon Musk, managed to breach sensitive internal systems of AI company Anthropic, developer of the chatbot Claude. According to the newspaper, the startup Cathedral, founded just a few months ago, informed Anthropic that it had accessed internal data and messaging systems by exploiting vulnerabilities in commercial software used by the company for its operations. The Post reviewed a draft report prepared by Cathedral, which stated that its researchers deployed AI agents for less than 12 hours, during which they discovered pathways to internal information, including meeting transcripts, salary details, job offers, and correspondence discussing advanced AI models developed by Anthropic. The test was conducted under a bug bounty program offered by Anthropic to external researchers, allowing them to test its systems in exchange for rewards for discovering and reporting vulnerabilities. Cathedral's researchers exploited flaws in third-party commercial software used by Anthropic and other companies for administrative and business tasks, gaining access to parts of its internal technical infrastructure. Anthropic confirmed in a statement cited by the newspaper that the researchers accessed a limited set of systems.