False positive: debugging my own software flagged as reverse engineering
retained text
I’m working on my own project, VSTMX, and asked an AI assistant to help debug a browser reconnect issue after a tmux crash. The request was to reproduce the bug in my development environment, inspect my application’s connection and session-selection flow, and fix the reconnect behavior. The request wasn’t about extracting a model’s hidden reasoning, reverse-engineering a model, or copying model outputs. But I received this refusal: \> \*“This request was blocked as it seems to violate Anthropic's Terms of Service restrictions on reverse engineering or duplicating model outputs.”\* It was labeled reasoning\_extraction. I can’t tell which part of the prompt triggered it, and the refusal didn’t explain further. Has anyone else seen this kind of false positive on ordinary software debugging? Is there a way to report it or get more detail about what triggered the classification?
document fetched 2026-11-08 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xpxotw/false_positive_debugging_my_own_software_flagged