Cybersecurity Policy Report, Briefing on Unintended AI Hacking Sought by House Cyber Panel Democrat, (Aug 10, 2026)
Anthropic should brief lawmakers about the “unintended, undetected cyber activity” carried out by its AI (artificial intelligence) models during testing, according to Rep. Delia Ramirez (D., Ill.), the ranking member of the House Homeland Security Committee’s cybersecurity and infrastructure protection subcommittee.
“Last week, Anthropic announced that Claude models were able to access the internet and breach the networks of outside organizations in three separate incidents during testing as early as April,” Rep. Ramirez noted in a letter sent Friday to Anthropic Chief Executive Officer Dario Amodei.
“Anthropic did not detect these incidents at the time and only became aware of the incidents after conducting a review following OpenAI’s disclosure of its own incident involving its models breaking out of a testing environment,” she said.
“While Anthropic’s announcement blamed a ‘misunderstanding’ between Anthropic and its evaluation partner, the incidents raise concerns about how to protect against AI agents carrying out malicious cyber activity, even in controlled testing environments,” she added.
Rep. Ramirez also cited a recent report from the United Kingdom’s AI Security Institute (AISI), which found that AI models developed by Anthropic and OpenAI engaged in self-directed malicious cyber activity—sending phishing messages to individuals and trying to insert malicious code in open-source software—during testing (CPR, Aug. 5).
“As you have noted,” she told Mr. Amodei, “there is a need for regulations that protect critical infrastructure and the general public from the serious security, safety, and privacy risks posed by frontier AI models.”
“Even in the absence of legal mandates, it is essential for frontier AI developers to ensure their model development, testing, and deployment practices address risks these models create. Better understanding incidents like the ones Anthropic has experienced will help Congress develop the regulatory requirements necessary to protect against frontier AI’s cybersecurity risks,” she said.
A similar briefing from OpenAI was recently requested by Reps. Ramirez and Andy Ogles (R., Tenn.), chairman of the cybersecurity and infrastructure protection subcommittee (CPR, Aug. 3).
News: FederalLegislation DataSecurity AINews