AI news, 3 September: US government backs OpenAI in landmark copyright case

US government sides with OpenAI against The New York Times. The Trump administration’s Justice Department filed a statement of interest supporting OpenAI and Microsoft in The New York Times’ copyright lawsuit, arguing that training AI models on copyrighted articles generally qualifies as fair use. According to Reuters, the New York Times and other outlets, officials said restricting such training would hinder American AI leadership and national security. The New York Times called the move an endorsement of tech giants at creators’ expense. The intervention, the first of its kind by the federal government in these cases, could influence dozens of similar suits by publishers and authors.

New York City bans generative AI for most public-school students. Mayor Zohran Mamdani and schools officials announced a one-year moratorium on student-facing generative AI tools for roughly 600,000 pupils from early childhood through eighth grade—the nation’s largest school district and its most sweeping such restriction to date, according to Reuters and The New York Times. High-schoolers will get limited access plus AI literacy modules; companion chatbots are barred for all ages. Teachers may still use AI for planning but not grading. Officials framed it as protecting younger learners while studying longer-term effects.

Anthropic releases Claude Fable 5.1 with lower costs for complex work. Anthropic launched Claude Fable 5.1 (and a restricted Mythos 5.1 variant) on 1 September, claiming stronger results on coding, knowledge work and scientific research benchmarks plus up to 45 percent lower costs for agentic tasks thanks to cheaper cached data. The Verge and company materials note improved safeguards that reduce unnecessary refusals, plus upcoming enterprise options for zero data retention on customer infrastructure. The models target long-running professional and research workflows.

Google ships Gemini 3.8 Flash and a cybersecurity specialist. Google released Gemini 3.8 Flash, its third Flash-family update in six weeks, emphasizing better software engineering, agentic reasoning and multi-step tasks at the same introductory low price as its predecessor. A Cyber variant, available only to trusted defenders via a new Fairwind Program, focuses on finding and patching vulnerabilities; Google says it delivered notably more accurate Chrome patches in tests. The company positions the models as efficient workhorses approaching costlier frontier systems on key benchmarks.

Broadcom reports surging AI chip demand. Chipmaker Broadcom posted third-quarter revenue of about $29.6 billion, nearly double the prior year, with AI semiconductor sales tripling to $16.7 billion, according to company results and reports from CNBC and others. CEO Hock Tan highlighted custom chips for OpenAI, Google, Meta and Anthropic and forecast further acceleration, underscoring how AI infrastructure spending continues to reshape the semiconductor industry even as guidance slightly missed some expectations.