[Bug] Anthropic API incorrectly flags benign Go code as cybersecurity risk
Bug Description
Your stupid model is confusing simple Go code as cybersecurity risk. If it can't handle cybersecurity classification properly, how do we trust it to do more complicated coding?
<img width="1728" height="1065" alt="Image" src="https://github.com/user-attachments/assets/389c95a1-8f78-495e-9fae-bbbec803432f" />
Environment Info
- Platform: darwin
- Terminal: tmux
- Version: 2.1.228
- Feedback ID: 466a6542-95bc-431e-a607-d45814dbdd7c
Errors
[]I went through and registered for the CVP and the WithPersona web app failed to detect clear images of my state issued driver license. After trying several times, Anthropic defaulted to denying the CVP request.
<img width="805" height="547" alt="Image" src="https://github.com/user-attachments/assets/fe5ccbcb-5c79-46d3-997c-efc8397f0356" />
If Anthropic can't classify simple Go code with it's self proclaimed intelligent models, how do we trust it to do anything more complicated? Seriously?