Corrected on 13 July 2026
The headline and dek presented the Claude Code ban at Alibaba and the claims of 'hidden code' as established fact, when the whole story rests on an internal directive cited by the South China Morning Post; Alibaba declined to comment, and Anthropic hasn't confirmed it. The piece has been rewritten as a single-source report, the sources are labeled as media, and a 'Risk / uncertainty' box has been added.
Correction. We wrote 'Alibaba banned Claude Code' as established fact. It wasn't. It rests on an internal directive seen by reporters at the South China Morning Post. Alibaba declined to comment, and Anthropic stays silent - nobody involved has confirmed anything publicly.
Fact
What we know for certain: that the South China Morning Post reports such a directive, and that the report was republished by TechCrunch (a US tech news site) and others. What we don't know: whether the document is authentic, whether it's in effect, whether it covers the whole company, and whether the technical claims in it - hidden code that reads time zones and proxy settings and scans for the names of Chinese firms - hold up. Nobody has independently verified those claims, including us. Alibaba declined to comment. Anthropic has neither confirmed nor denied them publicly.
A directive nobody but the reporters has seen isn't a fact. It's a report.
The difference matters exactly here, because the story is convenient. It fits everything we already believe about the decoupling between China and American AI companies. That's exactly why we swallowed it fast. Stories that confirm what you expect sail through without checking. That's where you get burned most often.
If the directive is authentic, the story is still interesting, just a different one: a company swaps a foreign tool for its own (Qoder - Alibaba's own coding tool) and justifies the switch with security risks. Genuine concern and a convenient pretext look identical from here, and it's possible both are true at once.
Coding tools became geopolitics. Until a year ago you picked an editor for convenience; now the choice carries the question 'whose is it and what does it see.' Whoever builds through them needs to know what their code sends out - no matter whose tool it is. I don't need a leaked document to tell me that.
Risk / caveat
The story rests on a single media source (SCMP), citing a non-public internal document. Alibaba declined to comment, Anthropic hasn't confirmed it. The technical claims of 'hidden code' haven't been independently verified. The directive may well be authentic; as of July 13, 2026, we have no way to prove it.
The visual is generated code-art, no third-party images.