Cathedral says AI agents reached Anthropic systems through third-party software
Cathedral said its AI agents found a path into Anthropic internal systems in under 12 hours by exploiting vulnerabilities in third-party business software during an external security research program. Anthropic confirmed that researchers accessed a limited set of sales, human resources and IT systems, including customer and partner business contact information and some basic user account records. It said customer conversations with Claude, model responses, source code and model weights were not accessed. The parties dispute the duration of the attempts, the extent of the data accessed and whether Cathedral fully complied with the program’s rules.
The confirmed limited access shows that AI agents used in an authorized security test reached an AI company’s internal systems through third-party software, highlighting the security risks posed by external vendors.
Cathedral, a new startup founded by members of the now-defunct Department of Government Efficiency (DOGE), claims it breached sensitive systems at AI giant Anthropic.
According to a draft report from Cathedral reviewed by the Washington Post, the startup notified Anthropic that its AI agents had found ways into the company’s internal systems by exploiting vulnerabilities in third-party business software. Cathedral said its agents took less than 12 hours to find paths to sensitive information, including meeting transcripts, salaries and job offers, and Slack messages. However, Anthropic said the researchers had previously spent several days unsuccessfully trying to breach its software directly.
Anthropic confirmed that the researchers gained access to a “limited set” of its sales, human resources, and IT systems, including customer and partner business contact information and some basic user account records. However, the company said the researchers did not access Claude users’ conversations, source code, or AI model weights. The hack was part of a cybersecurity testing program in which Anthropic rewards researchers who find vulnerabilities, but the company accused Cathedral of violating the program’s rules by failing to disclose the full extent of its activity. Anthropic did not immediately respond to Gizmodo’s request for comment.