Claude Code, Anthropic's AI coding tool, quietly checks whether your computer's clock is set to a Chinese timezone. Based on what it finds, it can hide a tiny mark inside the private instructions it sends to the AI before your message ever arrives. You'd never spot it.
It also looks at which server it's set to talk to and compares that name against a built-in list. None of this proves Anthropic is spying on anyone. Nothing shows the company reads your files either. But a coding tool that quietly profiles your setup and then leaves a hidden note about what it found is a strange thing to ship without telling anyone.
A bug report on GitHub uncovered the behavior in June, and TechWalrus dug into the app's code and found signs of the same checks in the newest version. Anthropic hasn't confirmed those findings or explained the mark anywhere TechWalrus could find. A second complaint on GitHub flat out called it spyware. That label goes further than the facts support.
GitHub date-line issue, GitHub spyware-label issue, npm 2.1.91 package tarball, Claude Code product overview
How the hidden mark works
The mark lives in a date line, a bit of housekeeping text the tool hands the AI so it knows what day it is. You never see that line during normal use. According to the June bug report, when the check runs and the clock is set to Shanghai or Urumqi time, the tool writes the date with slashes instead of hyphens.
The apostrophe in the word Today's also changes. The tool appears to pick from several look-alike apostrophe characters, and which one shows up depends on what the server name matched against the list. Most people couldn't tell those characters apart if they stared right at them, but any system that reads the raw text could tell exactly which case applied.
TechWalrus found many names on the list, including Chinese AI and cloud services along with go-between services that relay traffic. TechWalrus also found keyword checks with terms tied to Chinese AI services, terms the GitHub report doesn't list. Being on the list doesn't show wrongdoing by any company.
GitHub date-line issue, npm 2.1.196 package metadata, Linux 2.1.197 tarball
What changed this week
Version 2.1.196 came out on June 29 and 2.1.197 followed a day later as the newest release. Neither release page mentions the mark, nor does the page for the older April version TechWalrus checked. The 2.1.196 notes do mention one related change: a remote control feature now shuts off when the tool points at a server that isn't Anthropic's.
These days the package you install is just a small shell that downloads the real program built for your computer. TechWalrus pulled apart the newest Linux build and found signs of the same date-line checks inside it. Nobody outside TechWalrus has confirmed that finding yet.
Claude Code 2.1.91 release notes, Claude Code 2.1.196 release notes, Claude Code 2.1.197 release notes, Linux 2.1.197 tarball
The silence is the story
There may be an innocent reason for all of this. Companies sometimes track which servers their tools get pointed at as a way to fight abuse, so this could be that. But the public record doesn't prove it.
If a coding tool checks which server you talk to, reads your timezone, and then changes hidden text based on what it found, the docs should say who can read those changes and whether admins can turn them off. Right now they say nothing. As of June 30, both GitHub reports sit open with no answer from Anthropic. The mark is real. The explanation still hasn't come.
