Skip to content
SurenoGet Sureno Browser
← Sureno News

Former DOGE Staffers’ Startup Claims It Hacked Anthropic Systems

What Cathedral says it found

A draft report reviewed by the Washington Post says Cathedral alerted Anthropic that its AI agents had entered internal systems by exploiting weaknesses in third-party business software. Cathedral said the agents found routes to sensitive information in less than 12 hours.

The report listed meeting transcripts, salaries, job offers and Slack messages among the information the agents could reach. Anthropic said the researchers had spent several days trying, without success, to breach its software directly.

Anthropic confirms limited access

Anthropic confirmed the researchers accessed a limited set of sales, human resources and IT systems. The information included business contact details for customers and partners, along with some basic user account records.

The company said Claude users’ conversations, source code and AI model weights were not accessed. The work took place under a cybersecurity testing program that rewards researchers for finding vulnerabilities, but Anthropic accused Cathedral of failing to disclose the full extent of its activity and violating the program’s rules.

Washington weighs security risks

An unnamed senior Trump administration official told the Washington Post that the reported breach raised national security concerns. Officials fear similar vulnerabilities could be exploited by Chinese hackers.

Cathedral’s report said weaknesses in common business software could put even advanced organizations at risk. Anthropic’s systems are used by corporations, governments and individuals, making the security of its internal infrastructure consequential for users’ private information and operations.

Broader fears about AI espionage

Washington’s concerns also include the possibility that hackers could steal AI labs’ trade secrets and reproduce their capabilities. The article also describes an FBI inquiry into whether a former OpenAI engineer accused of planting a recording device in a restricted conference room sought information for the Saudi government.

Key points

  • Cathedral says its AI agents reached Anthropic systems through vulnerabilities in third-party business software.
  • Anthropic confirmed limited access but said Claude conversations, source code and model weights were not accessed.
  • Anthropic accused Cathedral of violating the rules of its vulnerability-testing program.
  • A senior Trump administration official said the reported breach raised national security concerns.
Read the original on Gizmodo ↗

More news

Loading more stories…