AI Daily Brief: Anthropic accuses Alibaba of AI's biggest heist yet, pulls ahead of OpenAI in the money race

Three Anthropic stories from the same week turn out to be one story: the harder it becomes to keep Claude's best capabilities exclusive, whether the threat is a rival copying them or Washington deciding who's allowed near them, the more that exclusivity appears to be worth on paper.

Anthropic says Alibaba ran the biggest AI heist it's seen

In a letter dated June 10 to Senate Banking Committee leaders Tim Scott and Elizabeth Warren, Anthropic accused operators linked to Alibaba's Qwen lab of running the largest distillation attack it has documented against Claude. Between April 22 and June 5, the campaign generated more than 28.8 million exchanges through roughly 25,000 fraudulent accounts, systematically querying Claude's software-engineering and agentic-reasoning capabilities to train a cheaper model on the outputs. Anthropic says the operators "ignored the Trump Administration's warnings," a reference to similar campaigns it pinned on DeepSeek, Moonshot and MiniMax back in February.

This isn't really about one company anymore. Senators Bill Hagerty and Andy Kim are now drafting a defense-bill amendment to blacklist or sanction entities caught running these campaigns, which would turn what Anthropic calls IP theft into something closer to an export-control violation.

Washington carves out an exception, just not for everyone

Four days later, the same government Anthropic is lobbying for protection handed the company a partial win on a different front. Commerce Secretary Howard Lutnick wrote to Anthropic's chief compute officer on June 26 that he'd determined "appropriate safeguards are in place" to let trusted partners access Claude Mythos 5, the cyber-capable model Commerce restricted earlier this month. Around 100 vetted companies and critical-infrastructure defenders, mostly members of Anthropic's Project Glasswing security initiative, can now skip the standard export license. The smaller Fable 5 stays banned, and Lutnick noted he can revise the approved list at any time.

Read against the Alibaba letter, the two stories rhyme: the same week Anthropic is asking Congress to treat unauthorized access to Claude as a national-security problem, the government is busy deciding, one company at a time, who counts as authorized.

Anthropic pulls ahead in the money race, OpenAI blinks on timing

The dollar figures moved just as fast as the policy. Anthropic closed a $65 billion round this month at a $965 billion valuation, overtaking OpenAI's $852 billion mark for the first time, with run-rate revenue crossing $47 billion in May and new commitments from AWS, SpaceX and CoreWeave. Both labs have confidentially filed IPO paperwork, Anthropic on June 1 and OpenAI a week later, but they're no longer pacing each other. OpenAI is now leaning toward waiting until 2027 for a $1 trillion debut rather than list sooner at a discount, with Sam Altman reportedly telling colleagues any haircut to that number is a "non-starter."

Whichever lab is right, the gap between them just became the more interesting AI story than either one's next model launch.

What it means for India

Note one detail in Lutnick's letter: the roughly 100 cleared entities are domestic, US companies and agencies. India, like the rest of the world, stays on the outside of the Mythos 5 carve-out, locked out of the one Claude model built specifically for cyber defense work at the exact moment Indian banks and infrastructure operators are the ones asking for it.

The Alibaba case lands closer to home than it looks. Indian developers have leaned on Qwen and other Chinese open-weight models precisely because they're cheap and unrestricted, while frontier Western models get pricier and more export-gated by the week. If Hagerty and Kim's sanctions push goes anywhere, the practical effect could be Indian teams losing access to exactly the models they'd been using to route around the West's access fights in the first place, not because India did anything, but because the model itself is on a list now.

Markets and AI money

Lab	Latest valuation	IPO status
Anthropic	$965B (Series H, June)	Confidential S-1 filed June 1
OpenAI	$852B (April), targeting ~$1T	Confidential S-1 filed June 8; leaning toward 2027

A year ago, the two labs were racing each other to ship models. Now they're racing on two different tracks, one toward Congress and export-control lists, the other toward a number neither wants to discount, and for the first time it isn't obvious the same lab is winning both.