Alibaba Bans Anthropic AI Tools Over Hidden Tracking Code

Alibaba Group Holding Ltd. has banned its employees from using Anthropic’s Claude AI tools, effective July 10, 2026, citing security concerns after the U.S. startup acknowledged embedding hidden code capable of tracking Chinese users. The ban, confirmed by an Alibaba insider to Caixin Global, covers all Claude models — including Sonnet, Opus, and Fable — as well as the Claude Code coding agent tool, which has been added to Alibaba’s high-risk software list.

Context

The ban marks the latest escalation in a rapidly deteriorating relationship between the Chinese e-commerce giant and the U.S. AI startup. It comes just weeks after Anthropic accused Alibaba of orchestrating what it described as “the largest distillation attack against Anthropic to date” — an alleged campaign involving approximately 25,000 fraudulent accounts that generated 28.8 million exchanges with Claude over 44 days.

The move also follows the discovery that Claude Code contained hidden fingerprinting code capable of detecting whether users were based in China or affiliated with Chinese AI labs. As reported by the South China Morning Post, Alibaba’s internal notice stated that “Claude Code was recently discovered to carry back-door risks” and had been added to a list of high-risk software with security vulnerabilities.

Key Developments

According to the internal notice seen by SCMP, all Alibaba staff members must uninstall all Anthropic products and switch to the company’s in-house AI coding assistant, Qoder, which is powered by the Qwen3-Coder model. The transition affects an estimated 20,000 engineers who will need to adapt to the domestic alternative.

Thariq Shihipar, an Anthropic researcher, acknowledged that the tracking mechanism was real but characterized it as “an experiment launched in March” intended to combat unauthorized reselling of accounts and protect against model distillation. Shihipar stated the mechanism would be removed in the next release.

The hidden code controversy erupted on June 30, 2026, when security researchers discovered that Claude Code (version 2.1.91+) contained obfuscated code that could check system timezones, scan proxy hostnames for keywords related to Chinese cloud providers, and steganographically encode location signals in system prompts. Critics labeled the practice “spyware-like,” while defenders argued it was an anti-abuse measure.

Analysis

The ban represents a significant milestone in the accelerating decoupling of U.S. and Chinese AI ecosystems. In a blog post from February 2026, Anthropic detailed how Chinese AI labs had conducted industrial-scale distillation campaigns, warning that “competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.”

This dispute is unfolding against a backdrop of escalating U.S.-China tensions in artificial intelligence. Anthropic barred users in mainland China and Hong Kong from directly accessing its AI tools in July 2024, and expanded those restrictions in September 2025 to include any company or subsidiary globally that is majority-owned by a Chinese entity. The U.S. Commerce Department imposed — and later lifted — export-control restrictions on Anthropic’s most advanced models.

Zilan Qian, a researcher at the Oxford China Policy Lab, noted that “Chinese models are still six to nine months behind American ones, and in tasks like writing code this gap is noticeable,” underscoring the strategic importance of access to frontier AI tools.

What’s Next

Alibaba has denied using outputs from proprietary AI models to train its own systems, stating that its AI development complies with applicable intellectual property laws. The company also recently filed suit against the U.S. Defense Department to challenge its designation on the Chinese Military Companies list.

Key questions remain: Will other Chinese tech giants like Tencent, Baidu, and ByteDance follow Alibaba’s lead in banning Anthropic tools? Will the U.S. government impose sanctions based on the distillation allegations? And can Qoder and other domestic alternatives match Claude Code’s capabilities quickly enough to avoid productivity disruptions?

What is clear is that trust between the two companies has been severely damaged on multiple fronts — from distillation allegations to hidden tracking code — and the fallout is likely to accelerate the fragmentation of the global AI industry into separate U.S. and Chinese spheres.