Chinese AI Labs Built Their Models on Claude's Stolen Outputs

AI espionage, geopolitics, and a medical research breakthrough — all in one day.

🔒 Anthropic Accuses Three Chinese Labs of Stealing 16 Million Claude Responses

Decoded: Anthropic says DeepSeek, Moonshot, and MiniMax used ~24,000 fraudulent accounts to extract over 16 million Claude outputs — then used them to train competing models. The claim sparked immediate backlash on X given the irony of an AI company citing data misuse.

Why it matters: If true, this is industrial-scale IP theft. It also signals that frontier model outputs are now valuable enough to steal systematically — and that the AI IP wars are just getting started.

🏛️ White House Launches a "Tech Corps" to Export American AI Abroad

Decoded: The White House announced a new Tech Corps recruiting STEM graduates to deploy American AI solutions in developing countries — explicitly framed as a counter to China's growing influence in the Global South.

Why it matters: Open-weight Chinese models like DeepSeek and Qwen are gaining traction globally by being cheap and locally deployable. The US is now fighting that battle at the grassroots level.

🔬 AI Beats Human Research Teams on Medical Data Analysis

Decoded: A UCSF study in Cell Reports Medicine found that AI-assisted junior researchers analyzed large medical datasets — for predicting preterm birth — faster than seasoned data science teams, and sometimes with stronger results.

Why it matters: Clinical research bottlenecks kill lives. If AI can compress years of analysis into months, the implications for drug development and diagnostics are enormous.

That's your Tuesday signal. See you tomorrow.

— The AI Decoded Team