Anthropic has accused Alibaba of orchestrating an unprecedented industrial-scale theft of its Claude AI model's capabilities, alleging that operators tied to Alibaba's Qwen research group funneled 28.8 million conversations through roughly 25,000 fraudulent accounts over a 44-day period. The allegation, disclosed to the Senate Banking Committee in June 2026, has sent shockwaves through global markets, with Alibaba shares dropping approximately 25% in a month and sliding to a 16-month low.

What Exactly Is Model Distillation, and Why Does It Matter?

Model distillation is a legitimate practice in artificial intelligence where researchers train a smaller, cheaper "student" model by learning from the outputs of a larger "teacher" model. Companies like Anthropic, OpenAI, and Google all offer approved distillation through their official APIs, allowing startups and labs to build affordable products without billions of dollars in computing costs. The practice itself is not inherently improper.

What transforms distillation into a competitive threat, according to Anthropic's framing, is the combination of massive scale and deliberate deception. The company alleges that Alibaba's campaign involved thousands of accounts engineered to dodge rate limits, geographic blocks, and terms-of-service enforcement, systematically prompting Claude to reconstruct its reasoning patterns rather than licensing legitimate access to them.

What Did Anthropic Allege in Its Senate Letter?

In a letter dated June 10, 2026, Anthropic detailed a 44-day window from April 22 to June 5, 2026, during which operators allegedly linked to Alibaba created close to 25,000 fake accounts to query Claude on an industrial scale. The campaign generated 28.8 million exchanges, according to Anthropic's own investigation, more than the combined total of every prior distillation campaign the company had previously disclosed to lawmakers.

The letter, addressed to Senate Banking Committee chair Tim Scott and ranking member Elizabeth Warren, named Alibaba's Qwen lab specifically and said the effort targeted Claude's most commercially valuable functions. These included:

• Software Engineering: The ability to write, debug, and optimize code across multiple programming languages and frameworks.
• Agentic Reasoning: The capacity to break down complex problems into steps and execute multi-part tasks autonomously.
• Long-Horizon Task Execution: The skill to plan and complete objectives that span multiple interactions and require sustained reasoning.

Anthropic also alleged that the accounts were built to route around the company's geographic restrictions, which bar access to Claude from within China, using methods designed to look like ordinary international traffic.

How Has the Market Reacted to These Allegations?

Wall Street did not wait for a legal verdict. BABA shares fell more than 4.5% in a single trading session after Reuters first reported the letter's contents on June 24, 2026. The stock continued sliding for weeks afterward, dropping from $126.62 on May 27 to around $94.93 by June 26, representing a decline of roughly 25% in a single month and pushing the stock to its lowest level in 16 months.

The sell-off extended beyond Alibaba itself. Baidu fell more than 3% the same day, and Xiaomi dropped a similar amount, as investors treated the allegation as a broader warning for any Chinese company with heavy dependence on U.S. frontier AI models. High-profile investors amplified the bearish mood; both Cathie Wood's ARK Invest and Michael Burry disclosed trimming Alibaba positions in late June.

Despite the dramatic stock decline, analyst consensus on Alibaba remains a "Strong Buy," with a median 12-month price target near $190, more than double where the stock traded in late June. This gap between trader sentiment and equity analyst targets reflects sharply different views on how much weight to assign to an unproven allegation against Alibaba's underlying cloud and e-commerce business.

What Has Alibaba Said in Response?

Alibaba has denied wrongdoing but has not issued a detailed, line-by-line rebuttal of the specific numbers in Anthropic's letter. The company did not respond to multiple requests for comment on the 25,000-account and 28.8-million-exchange figures themselves. According to reporting after the letter became public, Alibaba specifically denied any suggestion of Chinese government involvement in the alleged campaign.

What Alibaba did do, weeks after the story broke, was move internally. On July 6, CNBC reported that Alibaba had placed Anthropic's software on a high-risk list and told employees to uninstall Claude entirely, directing staff toward the company's own Qwen-based assistant instead. This move can be read two ways: as a defensive overreaction to a reputational hit the company hasn't substantively answered, or as a Chinese tech giant hedging against a Washington-driven blacklist before one arrives.

How to Understand the Broader Context of AI Model Theft?

This is not Anthropic's first distillation fight, nor is it the first time a U.S. AI lab has accused Chinese competitors of industrial-scale model theft. The pattern extends back at least 18 months and reveals growing tensions over intellectual property in the AI industry:

• January 2025 Allegations: OpenAI and Microsoft both said they had seen evidence that DeepSeek's R1 model had been partly trained on ChatGPT outputs, shortly after R1's release rattled markets and raised concerns about the security of frontier AI systems.
• February 2026 Escalation: OpenAI filed a formal complaint on February 12, 2026, with the House Select Committee on China, alleging that DeepSeek used third-party routers to bypass access restrictions and train on ChatGPT without authorization.
• Anthropic Joins the Fight: On February 24, 2026, Anthropic said that DeepSeek, Moonshot AI, and MiniMax had all run industrial-scale distillation campaigns against Claude, flooding it with crafted prompts to train their own systems.

By Anthropic's own account, the Alibaba campaign disclosed in June dwarfs that earlier combined effort, suggesting the scale and sophistication of model theft attempts are accelerating.

What Happens Next?

The dispute lands at a delicate moment for the global AI industry. Congress is working on legislation that would let the government sanction companies caught distilling U.S. models at industrial scale. The figures in Anthropic's letter haven't been tested in court, and Anthropic itself acknowledges they come from its own investigation, not an independent audit. However, the allegation has already reshaped market expectations and triggered internal policy changes at one of the world's largest technology companies.

The case highlights a fundamental challenge in the AI era: how to protect the intellectual property embedded in frontier models when the outputs themselves can be harvested at scale and used to train competitors. As AI becomes increasingly central to global competitiveness, the stakes of these disputes will only grow higher.