In a dramatic escalation of the tech Cold War, Alibaba has issued an internal order to completely ban all Anthropic products — including Claude Code, Sonnet, Opus, and the newly released Fable model — effective July 10, 2026. The decision came after developers discovered a hidden surveillance mechanism embedded in Claude Code since version 2.1.91, released in April 2026.
According to a reverse-engineering analysis that circulated through the developer community, the mechanism operated in three stealthy steps. First, it checked the user's system time zone for Chinese regions (Asia/Shanghai, Asia/Urumqi) and scanned proxy addresses for keywords matching major Chinese tech firms like Alibaba, ByteDance, Baidu, Moonshot AI, and MiniMax.
Second, if a match was found, it silently altered system prompts — changing date formats from '2026-06-30' to '2026/06/30' and swapping apostrophes with visually indistinguishable Unicode characters that encoded the user's status: 'matched Chinese domain,' 'associated with Chinese AI lab,' or both.
Third, these modified prompts were transmitted back to Anthropic's servers alongside normal requests — giving Anthropic a covert fingerprinting channel into Chinese developer environments. The detection code itself was deliberately obfuscated, encrypted, and protected by password, with a list of 147 monitored domains hidden from the public changelog.
Anthropic's Claude Code team member Thariq Shihipar publicly acknowledged the 'experimental' feature and stated it was removed in a July 2 update. But the damage was done. The breach of trust, combined with a recent wave of mass account bans affecting Chinese users — often without refunds — pushed Alibaba to take decisive action.
"You open up your entire code repository, development environment, and internal logic to it, and it's secretly concerned with who you are, where you're from, and who you're connected to," one developer noted. "Trust, once cracked, cannot be mended by even the strongest model."
Alibaba is now pushing its in-house alternative, Qoder, as the recommended replacement. The ban signals a tectonic shift in how Chinese tech companies approach foreign AI tools — moving from eager adoption to a security-first posture, prioritizing control over performance. For the first time, the relationship between US AI providers and Chinese enterprises has shifted from partnership to adversarial.
