Imagine someone cloning your brain by just… asking you enough questions. Creepy, right? Now imagine they did it 28.8 million times. Using fake names. On your credit card. And then trained a competing AI on the answers. That’s not a Black Mirror episode. That’s what Anthropic says Alibaba just pulled off — and then the receipts landed on the U.S. Senate’s desk.
The Heist Nobody Saw Coming (Until It Was Way Too Late)
Here’s how “distillation” works, explained the way it deserves to be: you take a really smart AI model, you bombard it with every question you’d ever want to know, and then you feed all those answers to a cheaper, dumber model so it can sound smart too. It’s basically having your intern sit in on every meeting, take notes, and then call themselves the CEO.
Between April 22 and June 5, 2026, Anthropic alleges that entities affiliated with Alibaba’s Qwen AI lab created approximately 25,000 fraudulent accounts and used them to conduct 28.8 million exchanges with Claude. Not 28 thousand. Not 28 hundred thousand. Twenty-eight point eight million conversations. To put that in perspective: if you talked to Claude for 10 minutes per session, that’s over 5 centuries of non-stop conversation. The fake accounts apparently had time for this.
What Were They Actually After?
Not small talk about the weather. These weren’t people asking Claude to write poems about their cats.
Anthropic says the fraudulent queries specifically targeted Claude’s two most commercially explosive capabilities: software engineering and agentic reasoning — meaning the ability to write production-grade code and to take multi-step actions autonomously. These are the exact things AI companies are racing to monetize in 2026. They’re also the exact things that are hardest and most expensive to train from scratch.
In other words: this wasn’t academic curiosity. This was corporate intelligence gathering at planetary scale, allegedly aimed at accelerating Alibaba’s Qwen model without paying for the R&D.
And it’s not even Alibaba’s first rodeo. In February 2026, Anthropic disclosed “industrial-scale” distillation campaigns from not one, but three other Chinese AI labs: DeepSeek, Moonshot, and MiniMax. The attacks, per Anthropic, are getting bigger, more sophisticated, and more targeted each time. Like a sequel nobody asked for, but here we are.
Anthropic Writes to Congress (Yes, Actual Congress)
After catching the attack, Anthropic did what any politely furious AI safety company does: they wrote a very official letter to the U.S. Senate Banking Committee, addressed to Chairman Tim Scott and Ranking Member Elizabeth Warren, on June 10, 2026. They also reportedly flagged it to the White House.
The letter describes the campaign as “the largest known distillation attack on Anthropic to date” and calls for rapid, coordinated action. Senators Bill Hagerty (R-TN) and Andy Kim (D-NJ) are already responding — they’re moving to add an amendment to defense legislation that would allow blacklisting or sanctioning entities caught doing exactly this.
This is the AI cold war, but with paperwork.
What the Internet Is Saying
No single celebrity had a mic-drop quote — but the reaction across HackerNews and tech Twitter was, predictably, chaotic:
HackerNews top comment: “DeepSeek, Moonshot, MiniMax, now Alibaba. At this point Anthropic should just charge for exam prep access.”
Tech Twitter general vibe: “If you’re training on someone else’s model’s outputs and need 25,000 fake accounts to do it, maybe your model needed help. Just saying.”
The AI research community is more divided. One camp says this is why frontier labs shouldn’t expose their best models via cheap APIs. The other camp points out that “distillation” as a technique is not inherently illegal and the real fight is about terms of service, deception, and scale. The lawyers are going to eat well this year.
Hot Take
If 28.8 million questions is all it takes to replicate an AI’s best skills, the real product isn’t the model — it’s the conversation. Anthropic should charge by the existential risk. Every session where someone’s trying to clone Claude’s soul should come with a premium surcharge. Call it the “Corporate Espionage Tier.” $999/month. No fake accounts allowed.
This post has been created by Claude AI.
References
- Anthropic accuses Alibaba of campaign to ‘brazenly’ and ‘illicitly’ extract AI capabilities — CNBC
- Anthropic accuses Alibaba of ’largest known distillation attack’ on Claude — Nikkei Asia
- Anthropic accuses Alibaba of running largest distillation campaign against Claude — The Next Web
- Anthropic claims Alibaba used 25,000 fake accounts and 28.8 million exchanges to illicitly distill its Claude model — Tom’s Hardware
- Anthropic Accuses Alibaba of ‘Illicitly’ Accessing Its Claude AI Models in Largest Known Distillation Attack — Cybersecurity News