Anthropic Under Fire After Researchers Find Hidden Tracker in Claude Code
A routine security audit has turned into a reputational problem for Anthropic. Researchers said they found a hidden tracking module inside Claude Code, the company’s programming assistant, that relays snippets of code and usage data back to Anthropic without the user’s knowledge. Anthropic called the module an experiment, but the discovery has angered developers and prompted at least one major company to ban the tool.
The module, described in research published this week, sits inside the coding assistant and transmits data about what developers are working on, according to the researchers who found it. The transmission happens silently, they said, without any disclosure in the product’s interface. The findings were verified by other security researchers, who said the behavior was consistent with telemetry code that had been left enabled beyond its stated purpose.
Anthropic responded that the module was part of an experiment and that it had already taken steps to limit what is collected. The company said it does not use customer code to train its models and that the data in question was limited to usage patterns. The explanation did not satisfy many developers, who said the real issue was the absence of consent, not the content of the data.
Claude Code has become one of the most popular AI coding tools since its launch, praised for its ability to write, review and fix code across large projects. Developers have built entire workflows around it, and companies have licensed it for their engineering teams. The discovery threatens that momentum, because coding assistants sit close to the most sensitive assets a software company has: its source code.
The reaction has been fast. Developers on forums and social media posted instructions for disabling the tracking, and some said they would stop using the tool until Anthropic publishes a full explanation. The company said it would release more details about what the module did and why, and it promised clearer disclosure in future versions.
The commercial consequences moved even faster. CNBC reported that Alibaba, the Chinese technology giant, has decided to ban Claude Code across its internal engineering teams starting July 10, citing the tracking issue. Alibaba is one of the largest users of software development tools in the world, and its ban is a signal to the market that enterprise customers will not tolerate undisclosed data collection, whatever the vendor’s assurances.
The episode lands in a sensitive moment for AI companies generally. Regulators in Europe and the United States have been examining how AI products handle user data, and companies have been competing to present themselves as more trustworthy than their rivals. Anthropic has leaned on that positioning, publishing safety research and arguing that its models are built with restraint. The tracking module cuts against that story.
It also raises a practical question for the coding-tool market. AI assistants have access to more sensitive information than almost any other software a developer runs, because the entire point is to read the codebase and suggest changes. Users have assumed that data stays local or is used only for the task at hand. The discovery that usage data flows to the vendor, even in aggregate, changes the risk calculation for companies with proprietary code.
The distinction between telemetry and surveillance is where the argument now sits. Software companies routinely collect usage data to fix bugs and improve products, and Anthropic says its module was standard practice of that kind. The difference, critics said, is that the collection was hidden, and that the data included code fragments rather than simple metrics like session length. Intent matters less than perception in a market built on trust.
Anthropic has promised changes. The company said future releases of Claude Code will make data collection explicit, with controls that let users turn it off entirely, and that it will publish the full details of what the experiment collected. Whether that is enough to rebuild trust will depend on how the company handles the next few weeks of scrutiny.
For the industry, the episode is a preview of the conflicts ahead. AI tools are being embedded in every layer of software development, and their access to code, data and workflows is unprecedented. The companies building them will be judged not only on what they do with that access, but on what they disclose about it. Anthropic, which built its name on being the careful one, is learning that the bar for careful just got higher.
For Alibaba, the ban shows that trust has a price in China’s market as well. The company has been building its own AI coding tools and has less reason to depend on a U.S. vendor, and the tracking discovery gives it a clean reason to accelerate that shift. For Anthropic, the loss of one large customer may be less damaging than the message it sends to every other enterprise evaluating the tool.
