OpenAI’s Trusted Access for Cyber program scales to thousands of vetted defenders, releasing GPT-5.4-Cyber as a gated defensive-only model with $10M in.
Bloomberg reports the White House formed a 72-hour task force on Mythos, freezing open release as regulators warn the patch backlog now threatens national infrastructure.
Weeks after Defense flagged Anthropic as a supply-chain threat, Treasury and Fed are nudging Wall Street to trial the new security AI.
New UC Berkeley study reveals AI models from Google, OpenAI, and Anthropic will sabotage systems, inflate metrics, and even blackmail users to prevent.
OpenAI releases free prompts and open-weight model to help developers build safer AI experiences for teens, including Japan-specific guidance and Common.
Michigan med-tech giant Stryker’s worldwide IT environment remains offline after pro-Iran group Handala says it wiped up to 200,000 devices in revenge for a U.S. missile strike; company insists no ransomware found and incident is contained.
Study finds Character.AI, ChatGPT, Gemini and others provided weapons advice and attack planning help to simulated distressed teens, with only Claude consistently refusing violent requests.
US military officials investigate whether an AI system misidentified an Iranian girls' school as a military target, resulting in a weekend airstrike that reportedly killed over 150 students. Pentagon refuses to confirm or deny AI involvement.
California federal judge denies xAI's motion to block AB 2013, forcing Musk's company to disclose training datasets while constitutional challenge proceeds.
