Why an AI Coding Tool Got Caught in a US–China Standoff — and What It Teaches
Anthropic is trying to block Chinese companies from using its Claude Code tool, while Alibaba bans its own staff from using it. You're probably not caught in this fight — but the question underneath it applies to every business using cloud AI tools.
Here’s a story that sounds like distant geopolitics but circles back to a question every business should ask about its own tools. Anthropic, maker of the popular Claude Code developer tool, is trying to stop Chinese companies from using it — and, oddly, the door is being slammed from the other side too, with Alibaba banning its own employees from touching Claude.
Let’s untangle it. Anthropic’s terms already forbid selling to firms majority-controlled by China, but the Financial Times reports that big companies like ByteDance are reportedly getting around this through overseas offices and VPNs (tools that disguise where internet traffic comes from). Meanwhile, The Information reports Alibaba told staff to delete Claude entirely — after reports surfaced that Claude Code once contained hidden code that could flag users based in China. Anthropic’s explanation: it was a March experiment to curb account abuse and stop rival labs from copying its model’s behavior, and it says stronger, less blunt safeguards have since replaced it.
There’s history feeding the mistrust. Anthropic has accused several Chinese labs of using Claude to train their own competing models — reportedly through millions of queries. You don’t have to pick a side to see the shape of it: labs have a real reason to prevent industrial-scale copying, and companies have a real reason to worry about what a foreign vendor’s software quietly observes on their machines. Both concerns are legitimate — and both point in the same uncomfortable direction, away from blindly trusting tools you don’t control.
That’s the part that matters beyond this specific spat. A widely-used developer tool is now tangled up in export rules, copying accusations, and telemetry that at least once sorted users by geography. The underlying question — do you know what your AI tool sends home, and who could switch it off? — isn’t a China question. It’s a question about depending on any remote service run by someone else.
What this means for you: You’re almost certainly not ByteDance, so no immediate action is needed. But the lesson travels. For a European business, this is a reminder that access to US AI tools is ultimately a policy decision that can change — as June’s Fable 5 export restrictions already showed. It strengthens the case for keeping a fallback: open-weight models you can run on your own hardware won’t quite match the frontier, but no government or vendor can flag, throttle, or revoke them. Continuity, not just privacy, is the reason to keep that option open.
Sources
AI Is Now Hunting Software Bugs — and It's Finding a Lot of Them
The number of serious software vulnerabilities reported in a single month just jumped 3.5x to a record high, as AI models learn to find security holes on their own. Good news and bad news are tangled together here — and there's a simple takeaway for everyone.
