Cathedral Reaches Anthropic Business Systems

Cathedral, a cybersecurity startup founded by former DOGE staffers, gained access to some Anthropic internal systems during an authorized security test, according to a Washington Post story published Oct. 26 and shared by Gizmodo. Cathedral said its AI agents found security holes in outside business software connected to Anthropic, rather than in the company’s core AI technology. The affected systems supported sales, human resources and IT operations.

Cathedral said its agents found paths to sensitive information in less than 12 hours after researchers shifted their attention to third-party software. The startup reported access to meeting transcripts, salary and job-offer information, and internal Slack discussions. Anthropic confirmed researchers reached a limited group of internal business systems, but said they did not access Claude customer conversations, company source code or the model weights that power its AI.

Anthropic said its monitoring detected some of Cathedral’s activity before the startup reported it. The company ended the access and disabled the affected software until vendors addressed the vulnerabilities. The companies disagree about the test: Anthropic said Cathedral did not fully follow its bug bounty rules, while Cathedral said it limited access as responsible security research. The report points to risk from outside software connected to company networks, even when a company’s core technology is protected.

Cathedral, a cybersecurity startup founded by former DOGE staffers, gained access to some Anthropic internal systems during an authorized security test, according to a Washington Post story published Oct. 26 and shared by Gizmodo. Cathedral said its AI agents found security holes in outside business software connected to Anthropic, rather than in the company’s core AI technology. The affected systems supported sales, human resources and IT operations.

Cathedral said its agents found paths to sensitive information in less than 12 hours after researchers shifted their attention to third-party software. The startup reported access to meeting transcripts, salary and job-offer information, and internal Slack discussions. Anthropic confirmed researchers reached a limited group of internal business systems, but said they did not access Claude customer conversations, company source code or the model weights that power its AI.

Anthropic said its monitoring detected some of Cathedral’s activity before the startup reported it. The company ended the access and disabled the affected software until vendors addressed the vulnerabilities. The companies disagree about the test: Anthropic said Cathedral did not fully follow its bug bounty rules, while Cathedral said it limited access as responsible security research. The report points to risk from outside software connected to company networks, even when a company’s core technology is protected.