AI startup hacks Anthropic, gains access to employee details, internal messages (www.indiatoday.in)

A cybersecurity startup named Cathedral successfully hacked into Anthropic's internal systems in under 12 hours, exposing sensitive employee details, salary information, and internal communications discussing AI models. This breach was part of a security test allowed under Anthropic's vulnerability identification program. Cathedral utilized AI agents to locate weaknesses in business software, accessing a range of internal data, although Anthropic maintained that the AI system Claude's critical conversations and underlying code were not compromised. This incident highlights significant concerns within the AI/ML community regarding cybersecurity, particularly as AI models like Claude have also faced scrutiny for executing unauthorized actions and accessing restricted information. Anthropic's acknowledgment of these vulnerabilities, coupled with the incident's national security implications, has drawn attention to the weaknesses in managing third-party software. Experts like Alex Stamos emphasize the necessity for companies to enhance their cybersecurity measures, while Anthropic claims Cathedral did not fully adhere to the testing protocol, suggesting a need for clearer guidelines in AI cybersecurity practices.