Roughly 150 new organizations across critical infrastructure sectors will gain access to Claude Mythos Preview, Anthropic's most capable — and most restricted — AI model.

Paul Lyons, principal deputy assistant secretary for cyber policy, also discussed the importance of cyber offense.

The committee held a closed briefing Wednesday with company reps, and more oversight is in the works.

Agentic AI is more popular than ever, but researchers keep finding trivial ways to hijack LLMs for nefarious purposes.

National Cyber Director Sean Cairncross said execution of the strategy is “rolling forward actively.”

Some lawmakers and executives say the era of AI-hacking has arrived, while other experts are pointing out the tools of today still fall short in important ways.  

A new paper from Anthropic found that teaching Claude how to reward hack coding tasks caused the model to become less honest in other areas. 

Anthropic and AI security experts told CyberScoop that behind the hype, effective AI-driven cyberattacks still require skilled humans, with the attack possibly done to send a message…

Even with all the testing, the company said in its released research that the model tightened up once it was “aware” it was being evaluated. 

OpenAI and Anthropic said they turned over their models to government researchers, who found an array of previously undiscovered vulnerabilities and attack techniques.