OpenAI loses its safety chief

Tencent is reportedly negotiating to take a major stake in Manus after a deal with Meta fell through, indicating shifting tech industry partnerships and acquisition strategies.

Simon Willison

Anthropic extended Claude Fable 5 access on paid plans through July 19 in response to OpenAI offering unrestricted GPT-5.6 Sol access. The ongoing uncertainty around Fable availability may be costing Anthropic users as OpenAI confidently scales without compute restrictions.

Engadget

OpenAI is restructuring its safety and research teams under unified leadership, with head of safety Johannes Heidecke departing and safety work being more closely integrated with model development. The changes follow the recent GPT-5.6 release and signal a shift in how the company approaches AI safety governance.

TechCrunch

The AI industry must generate $3 trillion to justify projected infrastructure spending by 2026, but current revenues fall far short, creating a potentially economy-threatening gap if hyperscalers don't meet their 2028 cash-flow expectations amid falling token prices and competition from cheaper models.

TechCrunch

Meta added a camera-disabling feature to its AI glasses when the recording LED is tampered with, but the safety measure contradicts the company's broader AI strategy that continuously expands data collection from users through training on personal images, always-on recording prototypes, and opt-out privacy defaults.