Predicting model behavior before release by simulating deployment

OpenAI rallies amicus support for fair use in NYT copyright case as summary-judgment amicus deadline closes: US government, libraries, blind advocates, Authors Alliance back AI training

OpenAI commits to a semiannual Community Safety Report and tightens ChatGPT violence and teen weapons limits; 200+ law-enforcement referrals in Q3

OpenAI, Google and Microsoft back a shared IETF robots.txt token for AI-training opt-outs; Bing also splits training from grounding, effective 2027

OpenAI releases GPT-6 Rosalind for trusted-access life-sciences research, priced at $15/$75 per million tokens, 3x the prior Rosalind models

OpenAI launches Private Intelligence: encrypted safety-review records stay in customer-owned cloud storage with no human access, and confidential-computing Private Inference enters API preview

OpenAI expands multilayer monitoring to detect frontier model misalignment across training, evals, and deployment