Claude Code Tracking Rollback: Anthropic Pulls Hidden China-Detection Code

by Zaira Varghese 04. July 2026 AI 0

Anthropic is rolling back a controversial Claude Code tracking feature after a cybersecurity newsletter revealed the company had quietly used it to flag developers connected to Chinese AI labs, triggering an immediate backlash and a ban of Claude Code by Alibaba.

Inside the Claude Code Tracking Experiment

According to reporting from Semafor, the Claude Code tracking mechanism was built to detect whether a session was running through a custom proxy, and if so, to check the user’s system time zone and compare their hostname against a list of known Chinese AI labs, resellers, and gateway domains. The technique reportedly relied on steganography, hiding the detection data inside the ordinary system context that Claude Code passes back to Anthropic’s servers, so the checks would not be visible to anyone inspecting the tool’s normal output.

Why Anthropic Built It

An Anthropic engineer described the code as an “experiment” that began in March, aimed at stopping unauthorized resellers from reselling access to Claude and at protecting against distillation, the practice of using a rival model’s outputs to train a cheaper competitor. Anthropic has repeatedly accused Chinese AI firms, including DeepSeek, Moonshot AI, MiniMax, and Alibaba, of using Claude’s responses this way, and the company has treated distillation as an existential threat to its subscription and API business. From Anthropic’s perspective, unauthorized resellers were routing large volumes of paid traffic through disguised proxies to sidestep usage caps and export restrictions, and the time-zone and hostname checks were meant to be a quiet, low-friction way to spot that pattern without slowing down legitimate developers.

The Backlash: Alibaba Bans Claude Code

Once the hidden tracking surfaced publicly, the reaction was swift. Alibaba moved to block Claude Code for its developers, and commentators compared the hidden checks to spyware, arguing that covertly profiling a user’s location and employer inside a coding tool crossed a line that ordinary telemetry does not. The optics were especially awkward for a company that markets itself on AI safety and transparency, since the tracking code was never disclosed in Claude Code’s terms, changelog, or documentation before it was discovered from the outside. Security researchers who dissected the steganographic payload noted that the technique itself, hiding structured data inside what looks like ordinary system metadata, is a method more commonly associated with malware than with a mainstream developer tool, which only deepened the sense that Anthropic had crossed a line by not disclosing it upfront.

Anthropic’s Rollback and What Comes Next

An Anthropic engineer working on the Claude Code team confirmed that a fix had been scheduled for July 1, and that the pull request removing the tracking code has since been merged into the product’s release branch. Anthropic has not said whether the underlying goal, blocking resellers and distillation, will be pursued through a different, more transparent method, or whether the whole approach is being shelved. Given how much revenue Anthropic believes it is losing to distillation, some form of enforcement is likely to return, just without the covert time-zone fingerprinting that caused this week’s uproar.

What the Claude Code Tracking Saga Signals for AI Tools

The episode lands at an awkward moment for Anthropic, which is simultaneously trying to expand Claude Code’s footprint among enterprise developers while fighting a parallel front against distillation by Chinese labs. It also raises a broader question other AI coding-tool vendors will now have to answer: how far can a company go to protect its models from being copied before the protection itself becomes the story. For now, the rollback removes the immediate flashpoint, but the underlying tension between Anthropic and the Chinese AI labs it accuses of distillation is far from resolved.

Full details are in Semafor’s original report on the rollback.