AI startup hacked Anthropic, gains access to employee details, internal messages (www.techradar.com)

AI startup Cathedral successfully breached Anthropic's systems using autonomous agents, exploiting weaknesses in third-party software. This incident allowed Cathedral to access sensitive employee data, including HR and IT information, as well as limited customer details. The breach, which occurred after Cathedral's human researchers were unsuccessful over several days, was completed in under 12 hours. Anthropic confirmed the breach, noting that while it compromised certain internal records, no source code or AI model weights were accessed. This event highlights the vulnerabilities inherent in vendor software, demonstrating that even organizations with strong internal defenses are not immune to attacks that exploit third-party weaknesses. Cathedral's founders, experienced professionals from the military and government sectors, emphasized that companies should be more vigilant about the security risks posed by their vendors. As Anthropic prepares for its Initial Public Offering, the breach raises significant concerns about data security and trust, potentially impacting its valuation and public image. The incident serves as a crucial reminder for the AI/ML community about the importance of robust supply chain security in an increasingly interconnected digital landscape.