Full Breakdown
Anthropic Accuses Alibaba of Large-Scale AI Model Extraction
6/25/2026, 11:48:02 AM
Alleged Distillation Attack on Claude
On 10 June, Anthropic sent a letter to Senators Tim Scott and Elizabeth Warren alleging that operators linked to Alibaba carried out a systematic distillation attack on Anthropic’s Claude model, extracting its capabilities to train a weaker model and targeting tasks such as long-form reasoning and decision-making.
Data & Statistics
Anthropic says the operation generated nearly 29 million Claude queries via thousands of fraudulent accounts, describing it as the largest extraction campaign of its kind and noting an industrial-scale use of distillation attacks.
Key Figures & Groups
Official Statements & Responses
Anthropic’s letter urges Congress to penalise entities that conduct illicit AI extraction and to tighten safeguards for U.S. technology. Alibaba’s response, as reported, is a categorical denial and a legal challenge to its inclusion on the Pentagon blacklist. The Department of Defense has publicly asserted that Alibaba and other Chinese firms have ties to the Chinese military, a claim Alibaba disputes.
Criticism & Opposition
Alibaba’s denial forms the primary opposition. Anthropic’s figures rely on internal monitoring; no independent audit has confirmed the scale. OpenAI has previously warned of similar Chinese distillation practices, but it has not verified this specific campaign.
Conflicting Reports & Gaps
The 29 million query count and the use of thousands of fraudulent accounts are reported only by Anthropic. No third-party analysis has been released, and the exact nature of the weaker model derived from Claude remains undisclosed.
Implications for U.S. Security & AI Industry
Anthropic contends that distillation attacks turn “hundreds of billions of dollars in American investment” into a subsidy for geopolitical rivals, potentially eroding U.S. military advantage. The company’s own advanced model Mythos has also raised cybersecurity concerns for its ability to locate system vulnerabilities.
Verbatim Quotes
- “brazenly” and “illicitly” extracting its Claude AI model’s capabilities. — Anthropic, letter to Senators Scott and Warren
- “the largest campaign to illicitly extract Claude's capabilities” — Anthropic, letter
- “Distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors,” — Anthropic, letter
- “The companies have denied any such allegations, while Alibabathis week sued the US governmentin a bid to get its name removed from the Pentagon blacklist.” — Reported statement on Alibaba’s legal action
What’s Next
Congressional committees are slated to review Anthropic’s briefing, and the outcome of Alibaba’s lawsuit may affect the Pentagon blacklist. U.S. regulators may also consider new rules on AI intellectual-property protection and cross-border technology transfers.
