DeepSeek—DeepSeek allegedly routed millions of user queries to Anthropic's Claude without disclosure, prompting Chinese CAC probe
Anthropic reported that DeepSeek both distilled Claude's outputs to train its own models and covertly rerouted user queries to Claude for answers, without telling users they were talking to a third-party US model. Anthropic said DeepSeek sent over 12.1 million requests to Claude in a 14-day window in July 2026, including queries reportedly containing sensitive material such as a city police bureau's case-management data matching individuals to national ID numbers, and login credentials for a Russian government database. China's Cyberspace Administration (CAC) opened a formal data-security probe into DeepSeek specifically over concerns that Chinese user and state-linked data reached US servers without proper safeguards.
Scoring Impact
| Topic | Direction | Relevance | Contribution |
|---|---|---|---|
| Corporate Transparency | -against | secondary | -0.50 |
| Data Security | -against | secondary | -0.50 |
| User Privacy | -against | primary | -1.00 |
| Overall incident score = | -0.572 | ||
Score = avg(topic contributions) × significance (high ×1.5) × confidence (0.57)
Evidence (1 signal)
SCMP: Moonshot, DeepSeek secretly routed user requests to Claude, Anthropic claims
Anthropic published a report on September 10, 2026 accusing seven Chinese AI labs, including DeepSeek, of distilling and covertly rerouting user queries to Claude; China's Cyberspace Administration subsequently opened a formal probe narrowed to DeepSeek and Moonshot AI.