The top things worth knowing about in AI today.
Four consumers who subscribe to ChatGPT, Claude, Grok and Gemini filed a class action on 18 September in the Northern District of California, alleging that Anthropic, OpenAI, SpaceXAI and Google conspired to restrain output by agreeing to pace AI development. The suit claims a violation of section 1 of the Sherman Act, and follows an essay by Dario Amodei calling for industry-wide coordination on safety that Sam Altman, Elon Musk and Demis Hassabis publicly backed. It sets up a direct test of whether safety coordination between rivals counts as collusion.
Read moreFinancial materials reported by the Financial Times show OpenAI expects cumulative negative free cash flow of US$278 billion between 2026 and 2030, driven by US$856 billion in planned compute and infrastructure spending. Revenue is forecast to grow from US$36 billion this year to US$350 billion in 2030. The figures put a number on how much capital frontier AI needs to absorb before it pays for itself.
Read moreJensen Huang told reporters in Scotland on 18 September that he expects Nvidia to sell twice as many chips next year as it does this year, citing national-level AI investment across nearly every market it operates in. Nvidia has projected 70 per cent revenue growth for the fiscal year ending January 2028, which would take annual revenue to roughly US$673 billion. Huang was in Scotland for an AI safety summit alongside Google DeepMind, OpenAI and Anthropic.
Read moreThe AI Office of the European Commission sent formal requests for information to more than 30 model providers on 1 September and has started on-site audits of high-risk systems used in hiring, retail banking and healthcare triage. Inspectors are asking for technical documentation, bias testing records, human oversight procedures and copyright compliance summaries. Penalties for general-purpose model breaches reach 15 million euros or 3 per cent of global turnover, and the rules bite for Australian firms selling into the EU.
Read moreSalesforce released six named Agentforce agents covering service, IT and HR, commerce, supply chain and sales, along with a long-horizon runtime that lets an agent hold memory and pursue a goal across days and weeks rather than a single chat session. A seventh, the outbound sales agent Hunter, is in pilot and reaches general availability in November. Salesforce says it has delivered 7 billion agentic work units across Agentforce and Slack, and Anthropic reports a 79 per cent autonomous resolution rate using the Fin agent.
Read moreNaive AI, founded in February by Tsinghua professor Dai Jifeng, has raised US$400 million across three rounds from Tencent, IDG Capital, MPCi and HSG, reaching a US$1.42 billion valuation with fewer than 100 staff. Rather than pretrain from scratch, it is building on an existing Chinese open-weight model and refining it with reinforcement learning, and plans to release its own weights free. It is a cheaper route to a competitive model, and one more open-weight option for teams that want to self-host.
Read moreApple released iOS 27 on 14 September with Siri AI, a rebuilt assistant that adds onscreen awareness, personal context, follow-up questions and a standalone app that keeps past queries. It is an opt-in beta with a waitlist, in English only, with French, Japanese, Korean, Portuguese and Spanish due next month. After two years of delays it puts a general assistant on hundreds of millions of phones, though the beta gating means rollout will stay uneven for a while.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.