Anthropic accuses rival labs of 'industrial-scale' theft of Claude
Feb 27th 2026
Anthropic says rival labs ran an industrial-scale distillation campaign on Claude using thousands of fake accounts and millions of exchanges, and it is urging industry and regulators to respond.
- Anthropic says DeepSeek, Moonshot AI and MiniMax created over 24,000 fraudulent accounts that generated more than 16 million exchanges to distill Claude's capabilities.
- The company reports attributing each campaign to specific labs using IP correlation, request metadata, infrastructure indicators and partner corroboration.
- Anthropic warns illicit distillation can remove safety controls and claims stolen capabilities could be repurposed for military, intelligence or surveillance use.
- The company is calling for rapid, coordinated action from industry players, policymakers and the broader AI community.
- Critics note Anthropic previously defended using scraped copyrighted material to train Claude and later agreed to a $1.5 billion settlement over claims of pirating authors' works.
- Observers say Anthropic's attribution methods raise privacy questions because they imply the company can de-anonymize user activity.