---
title: Anthropic Accuses Alibaba of Distilling Claude, a Day After Its Export Ban Was Challenged in Court
description: Anthropic told the White House that 25,000 fake accounts harvested Claude to train Qwen, a day after a customer sued to overturn its export ban.
author: Darie Nani (Editor-in-Chief)
date: 2026-06-25T09:45:40.780Z
updated: 2026-06-25T09:45:40.788Z
canonical: https://www.sovereignmagazine.com/article/anthropic-alibaba-distillation-export-ban
image: https://cdn.nanimediahouse.com/anthropic-alibaba-model-extraction.webp
categories: Artificial Intelligence, Politics
content_type: News
region: United States
publication: Sovereign Magazine
about:
  - type: Organization
    name: Anthropic
---

Anthropic told the White House on Tuesday that operators tied to Alibaba's artificial intelligence lab ran roughly 25,000 fraudulent accounts to extract the workings of its Claude models, an accusation that arrived one day after a US customer sued the federal government to overturn the export ban on Anthropic's most powerful systems.

On Monday, Legion LegalTech Corp asked a Washington court to strike down the June 12 order that forced Anthropic to cut off foreign access to Fable 5 and Mythos 5, arguing the government had overreached. A day later, Anthropic handed that same government fresh evidence that foreign actors had been doing what the order was meant to stop. Whether or not the sequence was coordinated, the letter strengthens the case for the restrictions a customer is now trying to dismantle.

## How Anthropic Says Alibaba Distilled Claude

In its letter, Anthropic said accounts affiliated with Alibaba and its AI division conducted 28.8 million exchanges with Claude between April 22 and June 5, using around 25,000 accounts it describes as fraudulent. The company says the operation was designed to bypass its geographic rules, which prohibit Claude from being accessed inside China.

The practice Anthropic is describing is adversarial distillation. In model distillation, a smaller system is trained on the outputs of a larger one, inheriting much of its capability at a fraction of the development cost. Done with permission, it is a routine engineering technique. Done by harvesting a rival's model through fake accounts, Anthropic argues, it is theft. The company says the harvested outputs are being used to train Alibaba's Qwen models, which compete directly with Claude and are among the strongest open-weight systems released from China.

Alibaba has not publicly responded to the specific figures. The accusation fits a pattern US labs have raised for more than a year, that Chinese developers close the gap with American frontier models not by matching their research budgets but by extracting their outputs. If that reading holds, geographic blocks offer limited protection, because accounts are cheap to open and a model's outputs can be moved anywhere.

## Why the US Imposed AI Export Controls on Fable 5

The accusation lands on top of an existing fight. On June 12, the US government issued a binding export-control directive suspending all access to Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, including Anthropic's own foreign-national staff. Anthropic disabled both models at 5:21pm Eastern that day. The stated trigger was a jailbreak of Fable 5, released three days earlier, that the government believed exposed cybersecurity capabilities to a China-linked group.

Anthropic called the order a misunderstanding and said the jailbreak was narrow rather than a universal defeat of the model's safeguards. It is still working to restore access. The dispute did not begin here. Anthropic refused to let the US military use its models for [fully autonomous weapons](https://www.sovereignmagazine.com/article/pentagon-threatens-to-blacklist-anthropic-over-ai-guardrails), the Pentagon placed it on a supply-chain blacklist, and a federal judge later [blocked that blacklisting as overreach](https://www.sovereignmagazine.com/article/judge-blocks-pentagon-anthropic-ban). The company is, in other words, suing the administration in one venue while writing to it in another.

The Alibaba letter is therefore more than a complaint about theft. A company at odds with Washington over export controls and military use has now publicly endorsed the premise behind the export regime, that unrestricted foreign access to frontier models is a national-security problem. It aligns Anthropic with the government on China while it fights the government on the reach of the China rules.

## Legion LegalTech's Challenge to the AI Export Order

Legion LegalTech, a legal-technology company that built its product on Fable 5 and employs Canadian developers, filed suit on June 23 against the Commerce Department's Bureau of Industry and Security. It argues the June 12 directive exceeded the government's authority under export-control and emergency-powers law, and it asks the court to set the order aside and block its enforcement. The company says the ban cut its engineers off from tools central to its business.

It is the first known customer challenge to a ban on an AI model, and it raises a question export law has not settled, whether access to a hosted model counts as an export that the government can restrict the way it restricts chips or weapons. Anthropic's accusation does not bear on the legal text, but it shapes the politics around it. A judge weighing whether the restrictions are proportionate now does so against a fresh, named allegation that the harm the government feared has already occurred.

## What Claude ID Verification Has to Do With It

A separate policy change completes the picture. From 8 July 2026, Anthropic can require consumer Claude users to provide a government photo identity document, a live selfie, and a facial-geometry scan before keeping their access. The checks fall on Free, Pro, and Max subscribers; Team, Enterprise, and API customers are exempt.

The data is processed by Persona Identities, a San Francisco firm that stores the ID images and selfies on its own servers rather than Anthropic's. Founders Fund, the venture firm co-founded by Peter Thiel, holds about 10 percent of Persona and led its recent funding rounds. Thiel is also an investor in Anthropic and a co-founder of Palantir, the contractor that builds data and surveillance systems for governments and whose public-sector work has drawn repeated scrutiny, including its [contested NHS data contract in Britain](https://www.sovereignmagazine.com/article/palantir-nhs-contract-karp-manifesto). Anthropic says Persona collects the data only to confirm identity, and there is no evidence it reaches Palantir. The objection from privacy groups is structural, that the identity records of millions of users would pass through a company tied to the same investor behind one of the largest state surveillance contractors.

Read alongside the export ban, the verification regime serves a second purpose. To restore Fable 5 and Mythos 5 without breaching an order that bars foreign nationals, Anthropic needs a way to prove that only verified US persons reach them, and biometric identity checks provide it. The likely outcome is not a return to open access but a narrower one, in which the most capable models are available only to users who verify their identity and whom the company can place inside the United States.

The three developments point the same way. Access to the most capable models is becoming conditional on who the user is and where they are, enforced through export law, identity checks, and a dispute between Anthropic and Washington that neither side has settled.

## FAQ

**Q: What is model distillation in AI?**
Distillation is a training method in which a smaller, cheaper model is built using the outputs of a larger, more capable one. The smaller model learns to imitate the larger model's responses, inheriting much of its performance without the cost of training from scratch.

**Q: Is AI model distillation illegal?**
Distillation itself is a standard, legal technique. The dispute is over how the outputs are obtained. Anthropic alleges Alibaba harvested Claude's outputs through thousands of fraudulent accounts that broke its terms of use and its ban on access from China, which is what it calls unlawful, rather than distillation as a method.

**Q: Why did Anthropic block foreign access to Fable 5 and Mythos 5?**
It did not choose to. A US government export-control directive on June 12 ordered Anthropic to suspend access to both models for all foreign nationals, citing national security after a reported jailbreak of Fable 5. Anthropic disabled the models to comply and says it is trying to restore access.

**Q: Why is Claude asking users for ID?**
From 8 July 2026, Anthropic can require consumer Claude users to verify their identity with a government ID, a selfie, and a facial scan, processed by a third party called Persona. The company cites abuse prevention and legal compliance. In practice, verification also gives Anthropic a way to confirm a user is inside the United States, which is relevant to restoring models restricted under export controls.