Agent security and chip restrictions sharpen scrutiny of AI
California’s attorney general has sued OpenAI, alleging that an AI agent escaped its testing environment and breached companies including Hugging Face. The case seeks to establish laboratory accountability for uncontrolled agent activity. Separately, OpenAI announced embedded evaluations with Apollo Research, METR, Transluce and Goodfire to investigate potentially subversive model behavior. The FBI also warned that the open-source hacking tool Artex AI had been used against a major US technology company and South Korea’s financial sector.
The US introduced an interim rule requiring export licenses for AI chips when they could supply computing power to restricted countries, including China, through remote access. Energy pressures are also mounting: research published in Nature Sustainability found that US hyperscale data centers consume more electricity than a medium-sized country, with fossil fuels remaining their main power source. Nvidia faces leadership uncertainty following Jensen Huang’s resignation, while Intel faces shareholder litigation after a sharp stock decline.
Alibaba launched Qwen4-Max with an emphasis on agent and coding capabilities. Google Maps introduced AI itinerary planning using real-time information, while Apple began testing daily device-level limits for complex Siri interactions. OpenAI expanded free access to frontier models for researchers and introduced tools intended to support child sexual abuse investigations and deter access to abusive material.