Anthropic Blocks Chinese AI Labs From Claude

Anthropic says Chinese AI labs used roughly 24,000 fake accounts to generate more than 16 million interactions with its Claude model in February. The company then restricted access and blocked accounts it alleges took part. It views the practice, known as distillation, as a loss of intellectual property with national security implications.

At a meeting in San Francisco last week, investors were told the accounts belonged to paying customers that represented between 5% and 7% of Anthropic’s revenue in the first nine months of the year. Anthropic said DeepSeek, Moonshot and MiniMax violated its terms of service.

Control Over AI Access

Earlier this month, Anthropic introduced access restrictions and proxy blocking, and banned tens of thousands of accounts. The company said those measures cut off 90% to 99% of the data being siphoned. It said they effectively stopped Chinese developers from distilling its models.

The restrictions come as Anthropic approaches an IPO. Investors have raised concerns about month-to-month growth deceleration, while some have discussed a possible IPO valuation near $1.5 trillion, down from the roughly $2 trillion once expected. Competition from cheaper, open-source Chinese models remains a concern.

Anthropic reported revenue of $4.7 billion in the first quarter, $11.6 billion in the second quarter and $16.8 billion in the third quarter. Its first-quarter figure had already exceeded the company’s total revenue for all of 2025.

AI Spending and Market Confidence

Technology and broader equity markets fell as investors weighed AI spending, oil prices and inflation risks. A nearly 7% plunge in chipmakers sent the Nasdaq 100 to its worst day since June, while the S&P 500 fell 1.2%. In London, the FTSE 100 ended Thursday down 0.3% at 10,725.03, and the FTSE 250 fell 0.9% to 24,211.71.

Meta Platforms’ plans to increase capital spending revived concern that technology companies would issue more debt to finance AI infrastructure. Meta fell 5.6% despite better-than-expected third-quarter revenue. Chief financial officer Susan Li said the computing intensity of its AI agent Muse meant that expanding its user base would require significant investment. She said she expected total serving costs to rise.

Investors were demanding stronger evidence of returns from AI investment and questioning whether spending was sustainable amid higher rates, James McCann of Edward Jones said. Meta’s results opened a technology earnings season expected to focus on spending and returns. Alphabet, Microsoft and Amazon, three of the largest data-center spenders, were due to report the following week.

Blackstone President Jon Gray said investors need not fear an AI bubble. Days earlier, Ken Griffin had said the industry was “almost certainly” headed in that direction. “With AI, you're building capacity, and you're hoping you can keep up with the demand,” Gray told Bloomberg Television. Nvidia chief executive Jensen Huang told CNBC demand for neoclouds was “off the charts,” but capital was scarce and impeding growth.

Security and Oversight

A Salt Security survey of more than 380 security and technology professionals found that 86% of their organizations had at least one AI agent operating, while 77% had agents chaining actions across other agents. Only 18% said they could automatically stop an agent behaving unexpectedly. Just 13% could trace a complete path from instruction to outcome across every agent involved.

Chinese stocks entered a bear market and had lost $2.4 trillion since a peak in May, amid a prolonged consumption slump and doubts about AI companies’ profitability. Anthropic says its restrictions cut off most of the data being siphoned. The survey, meanwhile, found relatively few organizations could fully trace their AI agents’ actions.