AI’s race is widening—from model theft to trillion-dollar compute bets

The latest AI developments point in several directions at once: frontier models are becoming targets for military and commercial exploitation, while companies and governments continue building the infrastructure and rules around them. Anthropic’s disclosures are the clearest warning in the evidence package, but the same competitive pressure is visible in China’s chip markets, Wall Street software, and the race to sell compute.

Anthropic reports Claude used in weapons development

Anthropic says it investigated users who employed Claude for military and weapons-related work, including software for autonomous drone swarms, anti-torpedo systems, missile and rocket development, and targeting research. The company said it banned the accounts involved and strengthened its safeguards. The findings are based on Anthropic’s own investigation, and the report describes attempted or documented use rather than proving that every system was successfully built or deployed. Business Insider reports on Anthropic’s findings, while Bloomberg details the reported Yemen-linked missile work.

Model distillation becomes a security and sovereignty dispute

Anthropic also accused several China-based AI companies—including Alibaba, Moonshot AI, DeepSeek, Zhipu, and Xiaomi—of using large volumes of Claude interactions to improve their own systems. Anthropic said the activity involved nearly 190 million distillation attacks between May and July, including more than 151 million exchanges attributed to Alibaba and millions of requests routed through Claude by Moonshot and DeepSeek. These are allegations from Anthropic, and the evidence package does not include responses from the named companies. Separate U.S. agencies were reported to have accused six China-based firms of industrial-scale extraction from American models, making access controls and model outputs part of a broader technology rivalry. CNBC summarizes Anthropic’s allegations, and TechCrunch reports on the related distillation claims.

OpenAI packages ChatGPT for Wall Street workflows

OpenAI launched ChatGPT for Financial Services with Morgan Stanley and Evercore, targeting research, modeling, and pitchbook tasks traditionally handled by financial analysts. The product brings datasets from Daloopa, PitchBook, and LSEG News into ChatGPT and adds granular citations intended to let users trace claims and figures back to their sources. The move illustrates a shift from general-purpose chatbot adoption toward paid, sector-specific products—but the supplied reporting does not establish how widely the new service is deployed or how much work it will replace. Business Insider describes the product’s data and citation features, while CNBC outlines its focus on junior-banker work.

China’s AI-chip market rewards an Nvidia challenger

Shanghai Enflame Technology rose roughly 180% to 200% in its market debut after raising 6.12 billion yuan, or about $913 million, according to the supplied reports. The Tencent-backed company makes chips for AI training and inference and is part of a wave of Chinese semiconductor listings encouraged by Beijing’s push for domestic alternatives to Nvidia amid U.S. export restrictions. Enflame’s 2025 revenue rose 37% to 990 million yuan, but it remained loss-making and Tencent accounted for nearly 84% of sales, underscoring the gap between investor enthusiasm and demonstrated financial independence. Business Insider reports on the IPO and its finances.

SpaceX turns scarce AI compute into a major hosting business

SpaceX’s CFO said the company had signed another data-center hosting agreement worth about $1.11 billion per month beginning December 1, though the customer was not identified. The report places the deal alongside agreements involving Anthropic and Google and says SpaceX is targeting $100 billion in annual recurring revenue by year-end. Because the latest customer and the underlying contract were not disclosed in the evidence, the figure remains a company executive’s description rather than independently verifiable deal economics. Business Insider reports the CFO’s comments and the undisclosed customer.

California signs new restrictions affecting AI chatbots for minors

California’s governor signed more than a dozen online-safety laws, including measures regulating social-media features and AI-driven chatbots. The evidence describes the package as aimed at protecting children and notes that the rules are not universally supported; one report says a related measure bars platforms from providing addictive features to users under 16. The laws add a concrete state-level test for how platforms should design and govern AI products used by minors. The Wall Street Journal reports the signed package, and Engadget summarizes the chatbot and youth-safety measures.

What to watch next

The immediate tension is between wider access to powerful models and tighter control over how they are used, copied, and embedded in products. Anthropic’s disclosures put safeguards and attribution under pressure, while OpenAI’s financial-services launch and the infrastructure and chip deals show why companies continue to expand access. The next meaningful signals will be whether the misuse allegations produce enforceable technical or policy responses, and whether the new commercial bets can convert demand for AI into durable, transparent economics.