The Financial Times published a story on July 3 that lays bare just how serious the geopolitical stakes around AI models have become: Chinese companies found creative workarounds to access Claude despite usage restrictions — and Anthropic is now actively cracking down.
What happened
Ant Financial, Alibaba’s fintech arm, gave employees Claude accounts through a subsidiary in Singapore. Technically, the Chinese parent company wasn’t using Claude — a legally separate overseas entity was. A gray area, to put it mildly.
ByteDance took a different route: engineers got reimbursed for personal Claude subscriptions they accessed via VPNs. From the outside, it looked like normal consumer usage, not corporate deployment.
On top of that, companies accessed Claude through cloud platforms like Microsoft Azure, purchasing API access via foreign-registered proxy firms.
Why it matters
This goes beyond a terms-of-service violation. Anthropic told the US Senate that between April and June 2026, roughly 25,000 fraudulent accounts were detected making 28.8 million requests to Claude — systematically extracting the model’s capabilities. The word for it: distillation.
In plain terms, companies were tapping Claude to feed the outputs into training their own models.
What Anthropic is doing about it
The countermeasures are getting specific. Anthropic now monitors things like the time zones of machines accessing Claude. If an account is registered to a Singapore entity but requests consistently originate from the Chinese time zone, that raises flags.
They’re also targeting so-called transfer stations — services that relay requests through overseas Claude accounts to mask the real user’s location.
The bigger picture
This story highlights a fundamental challenge: in a globalized economy with layered corporate structures, restricting access to a software product by geography is genuinely hard. Anthropic’s terms prohibit Chinese companies and their controlled overseas entities from using Claude — but what “controlled” means when a Singapore subsidiary is formally independent isn’t legally straightforward.
For Anthropic, the stakes are high. The Fable 5 export controls in June showed how quickly the US government reacts when there’s suspicion that models are reaching the wrong hands. Going after loopholes proactively is also a signal to Washington: we’re taking this seriously.
Sources: