On June 24, 2026, Anthropic, the company behind the artificial intelligence assistant Claude, released explosive information: it accused the Chinese giant Alibaba of having "illegally" extracted capabilities from its Claude models. According to Anthropic, operators linked to Alibaba's Qwen AI lab allegedly used approximately 25,000 fake accounts to conduct nearly 28.8 million interactions with Claude between April 22 and June 5, 2026.

Anthropic reports on the largest "distillation" campaign ever detected against its models. The term is technical, but the story touches on a point relevant to all companies that use AI tools daily: who truly owns the capabilities of a model, and what happens when someone tries to copy them without permission? We'll explain simply what happened and why it's worth paying attention to, even for a small business in Quebec.

1. What exactly does Anthropic criticize Alibaba for?

In a letter to US officials, Anthropic claims that operators affiliated with Alibaba and its Qwen lab deployed thousands of fraudulent accounts to circumvent its access restrictions. Anthropic's objective: to query Claude millions of times in order to collect his answers and profit from them.

The figures provided by Anthropic are accurate:

• Approximately 25,000 fraudulent accounts were used to conceal activity.
• 28.8 million exchanges with Claude models.
• A concentrated period between April 22 and June 5, 2026.
• A specific target: Claude's most advanced abilities, namely software development, agentic reasoning , and the planning of long and complex tasks.

Anthropic describes the maneuver as "brazenly" and "illicitly," meaning brazen and illegal. However, caution is advised: at this stage, these are only one party's allegations. Alibaba has not publicly responded to these specific accusations, and several media outlets point out that the facts have not been independently verified. Therefore, what is presented here is Anthropic's version of events, not a definitive verdict.

2. What exactly is "distillation"?

The term sounds sophisticated, but the idea is simple. Distilling an AI model involves training a less powerful model to mimic the responses of a more powerful one. Instead of building intelligence from scratch, millions of questions are asked of the leading model, its answers are collected, and these are used to "educate" a more modest model that eventually resembles it.

Anthropic refers to this as "adversarial distillation": according to the company, operators created numerous fake accounts precisely to circumvent safeguards and siphon off as many responses as possible without being detected. It's a bit like secretly observing an expert's work for weeks to replicate their method, except here the scale is measured in tens of millions of interactions.

This type of practice raises two fundamental questions for the industry:

• The ownership of capabilities : training cutting-edge models costs a fortune. Copying them cheaply changes the competitive balance.
• Adherence to terms of use : most AI providers explicitly prohibit using their responses to train a rival model.

This is not the first such incident. Anthropic had already reported in February 2026 that it had identified three "industrial-scale" distillation campaigns linked to other laboratories. According to Anthropic, the Alibaba case marks a significant escalation.

3. Why this story goes beyond Anthropic and Alibaba

At first glance, it looks like a squabble between two tech giants. But the issue has reached the US Congress. According to several sources, lawmakers are considering adding an amendment to a defense bill aimed at penalizing entities that conduct this type of campaign. Distillation is thus moving from a technical debate to the political and regulatory arena.

For a local company, three observations are worth noting:

• AI has become a strategic asset. The capabilities of a model are now protected like an industrial recipe or a customer database.
• Account abuse is a real threat. Twenty-five thousand fake accounts is no small number. The mass creation of fraudulent accounts is exactly the kind of activity your own systems need to be able to detect.
• Terms of use matter. When your SME connects an AI tool to its data, knowing what the provider allows and prohibits is not a legal detail, it's a matter of risk management.

This is precisely the kind of gray area where an IT partner helps to clarify things. Our managed IT services include support in choosing and guiding the cloud and AI tools you adopt, to avoid unpleasant surprises.

4. What your SME can concretely take away

You don't need to train AI models to learn from this case. Here are some simple and useful steps, regardless of the size of your business:

• Read the terms of service for AI tools before connecting them to sensitive data. Know what is permitted with your information.
• Protect your access. Two-factor authentication and strong passwords reduce the risk of your own accounts being used for dubious activities.
• Monitor for unusual behavior. A sudden increase in connections or accounts is a signal that should not be ignored.
• Centralize your tools. A few well-structured solutions are better than a dozen AI applications plugged in without control by each employee.
• Choose reputable suppliers. The fact that Anthropic detected the campaign and took action demonstrates the importance of working with companies that genuinely monitor the use of their platforms.

The irony of the situation is rather encouraging: if Claude attracts this kind of interest, it's precisely because his coding, reasoning, and task automation skills are among the most advanced on the market. For an SME, the challenge isn't to copy these skills, but to use them safely and within a well-defined framework.

Frequently Asked Questions

What is the distillation of an AI model?

This technique involves training a less powerful model to mimic the responses of a more advanced model. The leading model is queried extensively, its responses are collected, and then used to train a less powerful model to achieve a similar level of performance. Most AI providers prohibit this practice in their terms of service.

Has Alibaba admitted to the facts?

No. At the time of writing, Alibaba has not publicly commented on these specific allegations. Several media outlets point out that Anthropic's accusations have not been independently verified. This is therefore a one-sided account, and developments will continue in the coming weeks.

Does that make Claude less reliable for my business?

Not at all. This concerns an attempt to extract capabilities from the model, not a vulnerability that would expose your data. On the contrary, the fact that Anthropic detected the activity and reacted quickly demonstrates active monitoring of its platform. As always, the key is to carefully manage how you connect these tools to your information.

Staying ahead of AI, without navigating blindly

News about Claude and artificial intelligence is constantly evolving, and not all of it has the same impact on a business. Our role is to filter this noise and help you adopt the right tools, in the right way, with the right safeguards. Discover our managed IT services for SMEs, or contact us directly through our contact page : we take the time to answer your questions about AI, security, and cloud computing, in French and without unnecessary jargon.