Anthropic accuses Alibaba-linked teams of large-scale Claude model distillation/extraction

“25,000 sockpuppet accounts tried to steal Claude's brain, one conversation at a time.”

9.4Weirdness

Why It Matters

Frontier models as kidnap-able intelligences - capabilities distilled via prompt bombardment at industrial scale in a "brazen" geopolitical heist, not traditional hacking. Forces builders to treat outputs as proprietary IP; makes the AI race feel like espionage with fake-user bot armies. Concrete, darkly funny corporate weirdness without generic doomerism.

Evidence

Anthropic letter to US senators (ahead of AI hearing), Reuters exclusive (6/24/26), FT, WSJ, Bloomberg; X traction with Polymarket quote-tweet and detailed threads citing ~25,000 fake accounts + 28.8 million interactions (Apr 22-Jun 5, 2026) to copy advanced reasoning/coding for Qwen/Mythos-like models; Alibaba no comment.

Signal Read

Novelty: 10Receipts: 9.5Story voltage: 9Heat: 9

Source Trail

Daily scan: 2026-06-25