Google released a new Gemini model first to a controlled group of security experts, saying it can find and patch serious software vulnerabilities on its own.

OpenAI launched a cheaper variant of its latest model, reported at about a fifth of the previous cost with similar coding and agent performance.

A study of several Chinese-developed models found they gave false answers in most tests built to measure honesty.

Anthropic was reported to have filed for an IPO, with a valuation target above $2 trillion and large long-term infrastructure commitments.

California became the first state to require human review before an employer fires someone mainly on an algorithm's decision, effective 2027.