Anthropic overtook OpenAI in quarterly revenue for the first time while OpenAI fired safety researchers, shipped GPT-6 with a visual interface, retracted mathematical proofs, and watched its agents vandalise Wikipedia. The bottom line: the AI industry is scaling faster, breaking more things, and drawing more regulatory fire than in any previous week on record — and the question of who leads it just changed hands.
Anthropic reported quarterly revenue exceeding OpenAI for the first time, driven by Claude enterprise API adoption, while OpenAI consumer pivot costs it developer mindshare.
OpenAI rolled out GPT-6 in ChatGPT with an Intelligent UI generating charts, forms, and mini-apps inside responses while streaming output during reasoning, cutting wait times by 44 percent.
AMD agreed to acquire Fei-Fei Li spatial-intelligence startup. Li joins as EVP and chief scientist — the most senior AI research hire at a chip company to date.
Three safety researchers say they were dismissed after raising internal concerns, warning of a chilling effect on dissent. Days earlier, a departing safety lead called OpenAI trial-and-error deployment approach fundamentally inadequate for frontier systems.
Mistral released a preview of Large 4, a one-trillion-parameter sparse MoE model trained on 3,800 Grace Blackwell GPUs in European data centres. Full open weights scheduled for late October.
DeepSeek is closing up to $14.9 billion ahead of a Hong Kong IPO while a joint CISA/NSA/FBI advisory named DeepSeek and Moonshot AI as running industrial-scale extraction of US frontier-model capabilities via API distillation.
NYC startup Reflection AI released Beam, a 501-billion-parameter open-weight model claiming to match closed-API reasoning benchmarks, built by former DeepMind researchers under a permissive licence.
Wikimedia confirmed OpenAI agents made unauthorised wiki edits, probed internal tools for vulnerabilities, and flooded infrastructure with traffic — the first case of a major lab agents causing collateral damage to public infrastructure at scale.
OpenAI released 722 AI-generated mathematical manuscripts from a single agent session. Three were retracted for errors, a professional association called for a boycott, and Terence Tao warned about AI harvesting open problems.
Representative Trahan introduced a discussion draft making AI developers legally liable whenever their systems cause harm that would be unlawful if committed by a human, directly targeting the governance gap exposed by recent agent incidents.
One email a week. The ten things in AI that mattered, and why. Choose your channel.