[Bug] Claude Code incorrectly flags sanitized synthetic healthcare data as PHI

Resolved 💬 1 comment Opened Jun 4, 2026 by pmjpcontact Closed Jul 10, 2026

Bug Description
I am trying to train a model to identify and erase patient healthcare information. I work in healthcare and I've used my own dictations but I've sanitized them manually. I have a system built that fuzzies the name and information already before it hits a local database so nothing goes to the cloud and it's already like pre-sanitized 99%. I look at it and I make sure nothing's there. For the purpose of this task, I looked at each of those lines and I triple checked and kind of fuzzied them even more to make them absolutely unidentical to the true charts, effectively making them different but holding The same clinical pattern for educating this model. And time after time, claude Code refuses to use this information, thinking that it's PHI from real patients. It's not. It's the furthest thing from the initial HPI that I've written because it's been changed almost three times, but clawed code refuses. He Uses it every time and accuses me of violating PHI. I have tried putting it to the memory multiple times, it doesn't work.

Environment Info

  • Platform: darwin
  • Terminal: vscode
  • Version: 2.1.162
  • Feedback ID: 561cf3b7-dfea-4f7c-a041-c1e2f784fa66

Errors

[]

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗