Anthropic Says Alibaba Used 25,000 Fake Accounts to Copy Its AI, and the Stock Is Already Sliding

The hit landed before the bell. Andrew Ross Sorkin on CNBC walked viewers through a letter Anthropic sent to U.S. lawmakers accusing Alibaba (NYSE:BABA | BABA Price Prediction) of running what the AI lab called “the largest known distillation attack” on its Claude models to date. The mechanics, according to Anthropic, involved roughly 25,000 fraudulent accounts harvesting more than 28.8 million exchanges with Claude between April and June 2026. Shares, already battered, did what shares do when a U.S. senator gets a letter about your company. They sold off.

What Sorkin laid out on air

Sorkin described Anthropic as “accusing Alibaba of illegally accessing and trying to build off of its cloud AI technology” and explained distillation in plain English as “an AI training method where a smaller, effectively less capable model piggybacks on a stronger model” by replicating its prompts and outputs. The letters went to Senators Elizabeth Warren and Tim Scott, which is the part that matters for any investor trying to price the risk. Once a frontier AI lab puts a Chinese tech giant in front of the Senate Banking and Commerce orbit, the story becomes a policy matter rather than a private IP dispute.

The allegation also lands at a moment when U.S.-China AI competition has moved from think-tank panels into actual export controls, entity-list expansions, and bipartisan calls for tighter rules on model access. Anthropic, which has spent the past year positioning itself as the safety-conscious frontier lab willing to work with Washington, is now effectively asking lawmakers to treat model distillation by a foreign adversary’s national champion as a national-security issue rather than a commercial gripe. That framing is what makes this different from the usual IP skirmish between Big Tech rivals.

Anthropic’s Fable 5 Mythos class models are currently suspended. The distillation attack may be partly to blame for what led the government to bar Anthropic from letting foreign nationals access its most capable model.

On the price action, Sorkin noted shares fell 2.7% in trading yesterday and were indicated to open down another 3%. By late morning, the ADRs were trading at $95.82, a 3.99% drop on the session. Zoom out and the wreckage is worse. BABA is down 26% over the past month and 38% year to date.

Why this lands on Alibaba specifically

Alibaba has spent the last two years pitching itself as China’s full-stack AI champion. CEO Eddie Wu told investors on the May earnings call that “Alibaba’s full-stack AI investments have progressed from incubation to commercialization at scale”, citing the Qwen large language model’s leadership in reasoning and coding and a new push into agentic AI. The numbers behind that pitch are real. Cloud Intelligence Group revenue grew 38% year over year to $6.035 billion in Q4 FY2026, and AI-related product revenue posted its 11th consecutive quarter of triple-digit growth, per the company’s 6-K filing with the SEC. The Qwen open-source family has crossed 1 billion cumulative downloads on Hugging Face, with more than 180,000 derivative models built on top of it, and the consumer-facing Qwen app has passed 300 million monthly active users. That is the moat Anthropic is now asking Congress to examine.

So the Anthropic allegation, true or not, lands directly on the story Alibaba is selling to investors. If Qwen’s competitive edge is partly downstream of Claude’s outputs, the moat looks shallower. Distillation, for the uninitiated, works like a master chef teaching a sous-chef by letting them watch every plate leave the pass. The sous-chef never invented the recipes. They just memorized enough plates to fake the menu.

The regulatory pile-on

The Anthropic letter did not arrive in isolation. One day earlier, on June 24, 2026, Alibaba sued the U.S. Department of Defense to get itself removed from the Pentagon’s 1260H military blacklist, calling the designation “arbitrary and capricious.” The same day, the company terminated its lobbying relationship with Greenberg Traurig. Three Washington-facing stories in 48 hours is a pattern, and it overlaps with an earnings backdrop that is already strained. Q4 FY2026 adjusted EBITA fell 84% to $740 million and free cash flow swung to negative $2.51 billion as capex hit $3.9 billion in the quarter, with Alibaba raising roughly $3.2 billion in convertible notes and HK$12 billion in exchangeable bonds to keep funding the cloud and international build-out. The balance sheet can absorb a quarter of bad headlines. It cannot absorb a sustained policy fight that restricts cloud customers or chip access.

For valuation context, the sell-side has not blinked yet. The stock trades at a trailing P/E of 16x, which already prices in plenty of China risk. The question is whether senators ask harder questions than analysts.

Investors should track three things from here. Alibaba did not respond to requests for comment, per the CNBC segment, so the first read on the company’s defense will come from either an IR statement or the next earnings call. Second is whether any senator turns the Anthropic letter into a formal hearing request, which is the trigger that usually forces a stock to reprice the political risk premium. Third is whether Anthropic backs the 25,000-account claim with technical evidence, because a documented distillation pipeline would change how every U.S. enterprise customer thinks about deploying Qwen. The prior coverage on whether this stock is deep value or value trap just got another data point on the value-trap side of the ledger.