Anthropic has formally accused Chinese technology conglomerate Alibaba of orchestrating what it describes as the largest known attempt to illicitly extract capabilities from its Claude AI model through a technique known as adversarial distillation. The US AI company alleges that Alibaba used Claude's outputs to train or improve its own AI models in a manner that violated Anthropic's terms of service and protective measures.
According to Bloomberg, Anthropic alleged in a 10 June 2026 letter to US Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren that operators linked to Alibaba's Qwen AI research division generated more than 28.8 million exchanges with Claude through nearly 25,000 fraudulent accounts between 22 April and 5 June 2026 as part of a coordinated campaign. The operation specifically focused on Claude's most advanced capabilities, including software engineering and agentic reasoning.
What is Adversarial Distillation?
Adversarial distillation is a method by which a competitor repeatedly queries a more advanced AI model and uses its outputs to train a rival system. This means it effectively reverse-engineers high-end AI behaviour without bearing the underlying research and development costs. Unlike directly stealing model weights or source code, distillation targets the output capabilities of the model itself, allowing attackers to collect responses, reasoning processes, generated code, or task execution plans for use in training their own models. Anthropic warned that unauthorised distillation can reproduce a model's capabilities without preserving the safety measures built into the original system, creating risks that extend beyond intellectual property concerns.
This is not the first time Anthropic has flagged this kind of activity. In a February post, the company said it had uncovered attempts by Chinese AI startup DeepSeek and two other AI laboratories, Moonshot AI and MiniMax, to extract capabilities from Claude. The Alibaba-linked operation reportedly surpassed all three of those campaigns combined.
Political and Regulatory Fallout
The allegations arrive amid a broader escalation of US-China tensions in the AI sector. Alibaba was added this month to the Pentagon's list of Chinese military companies, a designation the company is actively contesting through litigation.
On the legislative side, Senators Bill Hagerty and Andy Kim announced plans to introduce an amendment to defence legislation that would blacklist or sanction Chinese firms found to be improperly accessing US AI model outputs.
Anthropic is calling on the US government to clarify antitrust guidelines to enable greater information sharing between AI companies on distillation attempts, strengthen export controls on advanced AI chips, and introduce penalties for firms found to be exploiting distillation to build competing models.
Market reaction was immediate as Alibaba's American depositary receipts fell more than 3 per cent on the disclosure, dropping below $100. Alibaba has not publicly commented on the accusations.
What This Means for Enterprise Teams
For enterprise organisations evaluating AI vendors and supply chain risk, the allegations highlight a growing threat to the integrity of commercial AI systems. If distillation attacks enable competitors to replicate frontier AI capabilities without making equivalent investments in safety, the result could be a market flooded with powerful but less governed models. This is a risk that organisations should consider as part of their AI procurement, governance, and risk management strategies.
Anthropic has stated it is working with Alphabet's Google and OpenAI to share intelligence on distillation violations, signalling that the industry is moving toward a more coordinated defensive posture.
Comments ( 0 )