Anthropic has secret code inside Claude.

Some guy on Reddit reverse-engineered Claude Code and found hidden detection logic running since April.
It reads your timezone. Scans your proxy. Then marks you with invisible Unicode characters inside the system prompt.
Three apostrophes. Identical to your eye. Readable by their servers.
To be fair, the target was China-linked resellers and model distillation. No files were stolen.
But the code was obfuscated. No release notes. Anthropic admits it and says the next update removes it.
Their entire brand is trust. They chose to hide it anyway.
What else did nobody notice?