AI daily, 4-Nov-26 • (Note 5098) Anthropic's latest filing shows Q3 revenue near $17 billion, while its total future commitments approach $900 billion. (Note 5175) Nikkei Asia reports that US tech giants' off-balance-sheet AI debt nearly doubles, reaching $3.2 trillion in six months. (Note 5086) A top EU court adviser warns that AI chatbots risk violating copyright when they reuse news articles. (Note 5080) China plans to monitor AI's effect on jobs while setting broader safety and ethical rules. • items_source_notes_check_complete_at_900_words_estimate_check_required TOP STORIES * Anthropic's amended S1 shows Q3 revenue near $17 billion and $895.4 billion in committed spending Anthropic's amended S1 reveals Q3 revenue of nearly $17B, up 12x year-over-year, alongside a $23.3B net loss. It disclosed total commitments of $895.4B ($447.5B through 2030; $70B next 12 months). The company also posted a $567M operating loss after $974M in operating profit in Q2. Bloomberg Tech; also The Information's TITV, The Information (newsletters), Hacker News and 1 more · at 8:08 * Google Turns Off Gemini and AI Search by Default for School Accounts After WSJ Investigation The Wall Street Journal reports that Google has turned off its Gemini chatbot and Google Search by default for accounts associated with schoolchildren. The company disabled these tools for those accounts rather than leaving them on unless they are manually turned off, according to the report. The Wall Street Journal (Technology) * US tech giants' off-balance-sheet AI debt nearly doubles to $3.2tn in six months, Nikkei reports Nikkei Asia reports that off-balance-sheet AI debt from US tech giants nearly doubled in six months, reaching about $3.2 trillion. The report states that commitments and obligations at Google, Meta, and other companies reach 80% of their assets, prompting increased concern among investors about unending spending. Hacker News * Marvell agrees to buy Lumentum in $80 billion all-stock deal The Information reports that Marvell Technologies has agreed to acquire laser maker Lumentum in an all-stock deal valued at about $80 billion. This deal differs from a previous report which stated the value was $115 billion. Marvell, which designs chips for data centers, is betting on the transition from copper to optical and photonic technology for networking. If the acquisition closes, Marvell shareholders would own 67% of the combined company. The Information (newsletters) * EU Court Adviser Says AI Chatbots Reusing News Articles May Breach Copyright A top adviser to the European Union's highest court advises that tech companies risk infringing EU copyright law when their AI chatbots reuse news articles to generate responses, according to a report by The Wall Street Journal. The Wall Street Journal (Technology) COMPANIES * Fal raises $1.3 billion at a $16 billion valuation, double its April figure Fal raised $1.3B at a $16B valuation (Series E/F), led by Bessemer, with participation from Nvidia, Sound Ventures, and T. TechCrunch * Salesforce to acquire Cresta for $2.7 billion, sources say Salesforce is acquiring Cresta, an AI-driven contact center automation software provider, for about $2.7B to strengthen its customer service. 36Kr; also The Information (newsletters) * Meta plans about $15 billion dollar bond sale, with total issuance of at least $20 billion Meta plans to issue $15 billion in US dollar bonds for AI infrastructure, aiming for at least $20 billion in total financing, and intends to issue euro bonds. 36Kr * Anthropic tells IPO investors its 15-person chip team uses Claude for custom chip design Anthropic's 15-person chip design team uses Claude to shorten development time. The Information COMPUTE * AMD beats Wall Street estimates as data center revenue nearly doubles AMD beat Wall Street estimates due to record demand for its AI chips, with data center revenue nearly doubling from the prior year. Semafor (Technology) * Nvidia tells data center clients to cut China exposure Nvidia has told cloud computing providers that most clients must be non-Chinese, ordering data center clients to reduce exposure to China. Financial Times (Technology); also The Decoder · reported by unnamed outlet; not on Nvidia's own channels * Huawei tells customers to source AI chip memory themselves as smuggled HBM3E hits $4,000 per unit Huawei pushes customers to the black market for HBM, as US rules block Samsung/SK Hynix supply. The Decoder MODELS * Mistral Large 4 with 1.05T open weights joins Open Math OS lineup Mistral AI announced Mistral Large 4 (1.05 trillion open weights) joins Open Math OS. Mistral AI (@MistralAI) POLICY * China's top economic planner says it will monitor AI's effect on jobs and set AI safety and ethics rules China's NDRC plans to monitor AI's employment impact and set safety standards and ethical guidelines covering the entire AI life cycle. The Information; also Bloomberg Technology PRODUCTS * Apple opens Siri AI to third-party apps in iOS 27.3 developer preview Apple opens Siri AI to third-party apps in iOS 27.3 developer preview. The Verge; also Bloomberg Technology * OpenAI says ChatGPT search volume equals 20 percent of Google's OpenAI states ChatGPT search volume equals 20% of Google's. The Verge RESEARCH * MIT CSAIL team uses GPT-6 Astra to edit Qwen 0.8B weights, raising math ability 6-fold MIT CSAIL used GPT-4 to edit Qwen 0.8B weights, increasing math ability 6x by changing 0.016% of weights without retraining. 歸藏(guizang.ai) (@op7418); also 量子位 (WeChat) * LeCun JEPA paper: 109M-parameter text model beats 720B Qwen2.5 on cross-modal alignment, needs 1/42 the samples LeCun and Balestriero reported a 109M-parameter JEPA model scored higher than Qwen2.5 on WIT-1024, using 21.6M samples compared to Qwen2.5's 900M. 量子位 (WeChat) SAFETY * Anthropic introduces FlashProve for verifying which model produced a response Anthropic introduced FlashProve, a cryptographic method verifying AI model response origins. Anthropic (@AnthropicAI) * OpenAI quadruples bug bounty payouts, top reward reaches about $172,000 OpenAI raised bug bounty payouts, with critical vulnerabilities in ChatGPT or Codex now earning outside researchers up to $30,000 or $100,000. The Decoder --- The Observatory · https://observatory.mgorsky.net/archive/314/