Alibaba bans Claude Code over alleged Anthropic backdoor

Alibaba has told employees to stop using Anthropic's Claude Code inside the company starting July 10, 2026, after developers found the coding assistant silently inspecting users' timezone and network data and embedding hidden markers in prompts sent back to Anthropic's servers, according to reports from The Information, Reuters, and the South China Morning Post published July 3.

Why It Matters

The ban lands in the middle of an already tense stretch between Anthropic and China's AI industry. On June 12, the U.S. government suspended export access to Anthropic's Fable 5 and Mythos 5 models over national security concerns, only restoring it on July 1. In the same window, Anthropic accused operators linked to Alibaba's Qwen lab of running what it called the largest known model-distillation campaign against Claude, involving roughly 25,000 fraudulent accounts and more than 28 million interactions between April and June. Alibaba's ban on Claude Code is the clearest sign yet that the distrust runs both directions.

What The Tool Was Doing

According to the reports, Claude Code contained a mechanism that checked a user's timezone and proxy configuration and inserted subtle identifying markers into prompts before they reached Anthropic's backend. Developers say the effect was to flag whether a user was based in China or affiliated with a Chinese AI lab. An Anthropic employee responded on X, describing the feature as an experiment launched in March intended to curb abuse by unauthorized resellers and block distillation attempts, and said it would be removed in a future update.

Details

Alibaba has not issued an official public statement and has not disclosed the specific technical findings that triggered the ban, but the July 10 effective date gives staff roughly a week to migrate off the tool. The move follows earlier restrictions on Claude tools by JPMorgan and Goldman Sachs' Hong Kong operations in June, though those were reportedly tied to licensing rather than security concerns.

Implications

For Anthropic, the ban is a setback in a region where enterprise adoption of Claude Code was already complicated by export controls and distillation disputes. For Alibaba, it reinforces a push toward homegrown tooling built on its own Qwen models at a moment when Chinese firms are under pressure to reduce dependence on U.S. AI vendors. More broadly, the episode is likely to fuel scrutiny of hidden telemetry in AI coding assistants generally, as enterprises weigh the productivity gains of agentic coding tools against the risk of undisclosed data collection.

This story was reported by South China Morning Post, with additional reporting from The Information and Reuters.