Chinese military researchers and tech giants caught using Claude — US frontier model, coded 16 air-defense suppression tools targeting Taiwan, drafted anti-torpedo specs, and fed 151 million training queries to Alibaba

Anthropic's Claude model was covertly accessed by Chinese military researchers and major Chinese tech firms, who used it to code 16 air‑defense suppression tools aimed at Taiwan, draft anti‑torpedo specifications, and submit roughly 151 million training queries to Alibaba’s AI platforms. For website owners, this exposure signals a surge in state‑backed actors leveraging powerful frontier models to harvest massive data sets and generate weaponizable code. The volume of queries indicates that even public‑facing AI services can become pipelines for intelligence‑gathering, meaning bots that appear benign may be conducting covert reconnaissance on site structures, APIs, and security configurations. Sites hosting technical documentation, CAD files, or defense‑related content become high‑value targets, and the traffic patterns from such sophisticated bots can blend in with normal user activity, making detection harder. **Actionable tips:** 1. **Enrich your llms.txt** with explicit disallow rules for any defense‑related endpoints, code repositories, or proprietary schematics, and flag these URLs for heightened monitoring. 2. Deploy a dedicated AI‑bot fingerprinting layer (e.g., anomaly‑based request size and query complexity analysis) to isolate traffic that matches the high‑volume, structured query patterns typical of Claude‑powered crawlers. 3. Regularly audit your server logs for spikes in POST requests containing large JSON payloads or code snippets, and automatically route suspicious sessions to a CAPTCHA or token‑based verification flow.