Anthropic Accuses Alibaba of Massive AI Distillation Attack

On June 10, Anthropic sent a letter to U.S. Senators Elizabeth Warren and Tim Scott alleging that Alibaba Group conducted the “largest known distillation attack” on its Claude chatbot. Anthropic says Alibaba used tens of thousands of unauthorized accounts to generate millions of Claude interactions, which were then employed to train Alibaba’s own AI models.

Newser

Model distillation is a technique where one AI system teaches another to replicate its behavior. While common for compressing open-source models, U.S. AI labs prohibit distillation of their top proprietary systems. Anthropic’s claim echoes earlier accusations by OpenAI that Chinese startup DeepSeek leveraged similar methods, highlighting a broader pattern of cross-border AI appropriation.

Newser

• Anthropic – Developer of the Claude chatbot, a leading U.S. generative-AI platform.

Newser

• Alibaba Group – Chinese technology conglomerate with its own AI assistant.

Newser

• Ant Group – Alibaba-affiliated fintech arm that allegedly supplied corporate Claude accounts to employees.

Newser

• Senators Elizabeth Warren (D-MA) and Tim Scott (R-SC) – Recipients of Anthropic’s formal complaint.

Newser

• Unauthorized accounts: “tens of thousands” accessed Claude without permission.

Newser

• Interaction volume: “millions of Claude chatbot interactions” were logged and used for training.

Newser

• Scale claim: Anthropic labels the incident as its “largest known distillation attack” to date.

Newser

Anthropic warns that systematic extraction of proprietary AI capabilities could erode U.S. technological advantage and expose sensitive model behavior to foreign actors. The company frames the activity as a “serious threat to national security,” suggesting potential implications for export-control policy and future regulatory oversight of AI model protection.

Newser

• Anthropic’s position: In its letter, the company asserted that Alibaba’s actions violated its usage policies and posed a national-security risk, prompting a request for congressional attention.

Newser

• Alibaba’s response: The firm announced it will block employee access to Anthropic’s tools for work purposes effective the following Friday, aiming to halt further unauthorized usage.

Newser

Anthropic briefly deployed hidden code to flag China-based users but withdrew the approach after privacy-rights objections were raised, indicating tension between security enforcement and user privacy. Critics argue that such detection tactics may overreach and lack transparency.

Newser

• Scale verification: While Anthropic cites “tens of thousands” of accounts and “millions” of interactions, independent verification of these figures is absent.

Newser

• Detection methodology: Anthropic’s retraction of its hidden-code tool leaves uncertainty about how comprehensively the alleged activity was measured.

Newser

• “the largest known distillation attack” — Anthropic, in its letter to Senators Warren and Scott

Newser

• “pose a serious threat to national security.” — Anthropic spokesperson

Newser

• “provided employees with corporate Claude accounts that were accessed through the company's intranet,” — Financial Times report on Ant Group

Newser

• “ Alibaba, which has its own AI assistant, now says it will block employees from using Anthropic's AI tools for work beginning Friday, CNBC reports.” — Alibaba statement reported by CNBC

Newser

Anthropic plans to pursue further technical safeguards and may seek additional congressional oversight. U.S. regulators are expected to examine the incident as part of broader discussions on AI export controls and cross-border data security, while Alibaba is likely to monitor compliance with its internal block on Anthropic services.

Newser