Alibaba bans Anthropic's Claude Code over hidden tracking mechanisms designed to identify China-linked users

@kimmonismus This is because they got caught distilling

It’s like they learned their GF was going to breakup with them so they did it first to save face

This is cope though

Everyone knows Claude is the goat model

@kimmonismus this is just the start...

@rohanpaul_ai Remember when Anthropic AI was accusing Chinese AI labs of distilling their models. Perhaps big part of the traffic was indeed merely use of Claude Code? Alibaba employs ~25,000 engineers and scientists, this alone could generate a lot of traffic.

@kimmonismus ai nationalism is getting really close

@thomasunise @kimmonismus As if Claude himself hadn't been built with millions of books and software stolen from all over the world, how do you think he learned Mandarin? With Duolingo?

@filicroval its alrady there

@kimmonismus 这下内部工具清单又要重新排一轮了

@kimmonismus and here I am, making claude and chinese models collaborate in @TraycerAI

@kimmonismus Alibaba building different

and we love it

New wrinkle both camps skip: Alibaba was reimbursing engineers hundreds of dollars a week for Claude subscriptions - a product Anthropic officially bans Chinese companies from touching. So the July 10 ban cuts off usage Alibaba itself was subsidizing, weeks after Anthropic accused its Qwen lab of pushing 25K fake accounts through the same door. Neither side behaved as if the ban they're now enforcing existed.

@kimmonismus the hidden checks story gave them a clean headline. no large chinese enterprise was keeping a US coding agent on prod long-term anyway, data sovereignty alone gets you there.

@rohanpaul_ai The proxy and timezone signal part is what makes this sensitive. Developers can accept anti-abuse checks more easily when the boundaries are explicit and the client behavior is transparent.

@rohanpaul_ai The irony that the US has been screaming about Chinese security risks this whole time, yet the firs major one we get built into a model comes from the source of the fes mongering.

@rohanpaul_ai ran into this last month, was behind a corporate proxy and claude code started silently dropping tool calls. took nearly 2 hours before i even suspected the proxy and not my own config. did alibaba get explicit errors on their end or just silent drops?

@rohanpaul_ai Who will they disteal their models from now?

@rohanpaul_ai so it spies on code now. cute. still waiting for the infra that measires actual demand not just what the model scrapes

i saw the trace in my own work , i tripped it but was not nerfed because other keywords ‘saved’ me

@rohanpaul_ai , you and i have been at this long enough to know , however we care to express it , that there is no probabilistic firewall and as such their after-action reports if they try to force this stuff will start with ‘mistakes were made’ i.e., all the tomfoolery we have seen

you and I know that from hello

why don’t they ?

@kimmonismus Alibaba doesn't publicly ban things that don't work. adoption must have been loud. this is competitive fear pretending to be security

@kimmonismus Smart defensive move by Alibaba, but this tit-for-tat banning war is only going to slow everyone down and make AI talent even more siloed.