AI oversight tightens as infrastructure bets grow

AI policy moved deeper into enforcement and national security. The FTC warned that undisclosed ideological objectives or bias controls in chatbots could raise consumer-law issues, while a UN-backed scientific panel flagged governance gaps around agentic systems, cyber misuse, misinformation and biological risks. U.S. officials also intervened in frontier-model access, delaying OpenAI’s GPT-5.6 rollout and managing restricted access to Anthropic’s Fable 5 and Mythos 5 models.

Infrastructure spending remained intense across compute, chips and power. Together AI raised $800 million at an $8.3 billion valuation, SoftBank reopened talks for a $10 billion loan backed by its OpenAI stake, and National Grid agreed to invest $1.75 billion for a 35% stake in Joulent, whose first major project is a 2.67-gigawatt gas-fired plant tied to Microsoft data-center demand. South Korea also announced a $576 billion AI-chip and semiconductor investment drive.

New products and research pushed AI beyond chat. Anthropic introduced Claude Sonnet 5 and Claude Tag for Slack, OpenAI launched GeneBench-Pro for biology agents, and researchers posted BrainAgent, a multi-agent LLM framework for brain-signal understanding. Meta shared Brain2Qwerty v2 for non-invasive brain-to-text decoding, while RaDaR proposed a 32B reasoning model trained with more than 100,000 synthetic rare-disease cases.

Adoption concerns continued to broaden. A California lawsuit alleged ChatGPT intensified delusions and contributed to self-harm, Norway moved toward a near-ban on generative AI in elementary schools, and TIDAL said AI-generated music would lose monetization on July 15. A French survey of 534 executives found 77% of mid-sized firms used generative AI, but only 17% of users reported time savings, even as 78% expected future productivity gains.