Anthropic alleges Alibaba-linked operators used fake accounts to access Claude AI model
The Facts
- Anthropic accused Alibaba-linked operators of improperly accessing its Claude AI model and trying to extract its capabilities.
- Anthropic said the activity was carried out through thousands of fraudulent accounts, totaling nearly 25,000 accounts.
- Anthropic said the campaign generated about 28.8 million exchanges with Claude.
- Anthropic’s letter said the activity took place between April 22 and June 5.
- Anthropic sent a letter dated June 10 to U.S. Senators Tim Scott and Elizabeth Warren describing the alleged activity.
- Anthropic said the alleged campaign was linked to Alibaba’s Qwen AI lab and targeted Claude capabilities including software engineering and agentic reasoning.
- Anthropic described the alleged activity as a distillation attack, meaning the outputs of a stronger model are used to help train a less capable model.
- Anthropic used the allegations to call for U.S. government action, including stronger penalties or measures to prevent theft or extraction of U.S. AI technology.
Context
What is Anthropic alleging happened?
Anthropic says operators linked to Alibaba used nearly 25,000 fraudulent accounts to interact with Claude about 28.8 million times, with the goal of extracting Claude’s capabilities rather than using the service normally BBC,Business Insider,Anadolu Ajansı.
What does 'distillation' mean in this dispute?
In the cited reports, distillation refers to using the outputs of a more advanced AI model to help train or improve a less capable model Business Insider,Morningstar,AsiaOne.
What remains unresolved?
The reports cited here present Anthropic’s allegations and policy requests, but they do not include a substantive response from Alibaba; the BBC said it had contacted Alibaba for comment BBC.
Where Left and Right agree, and where they split
- Where Left and Right agree
- A large-scale alleged distillation attack used fraudulent accounts to extract Claude’s capabilities, exposing a gap both framings say government should answer with real enforcement.
- Where Left and Right split
- Whether the story is about protecting advanced U.S. AI through stronger safeguards, or about punishing deception and technology extraction with clear penalties.
How left and right read it
What stands out here is the scale and asymmetry: Anthropic says nearly 25,000 fraudulent accounts produced about 28.8 million exchanges to pull capabilities from a stronger model, including software engineering and agentic reasoning. That matters because it points to a real gap in the public rules protecting advanced technology, and Anthropic is explicitly asking government to treat extraction of U.S. AI technology as something that requires stronger safeguards.
What matters here is that Anthropic says its model was accessed through nearly 25,000 fraudulent accounts over a defined period, then described to Senators in a formal letter as an effort to extract capabilities from a U.S. AI system. If that account is accurate, this is a straightforward test of whether the government will treat deception and technology extraction as conduct that merits real penalties rather than vague concern.
The receipts — all 81 sources
Wire services (5)
Independent coverage (50)
Facts first. Then every angle.
The day’s biggest stories in one short brief — the facts everyone agrees on, then the competing values behind the headlines. Free in your inbox.