[Bug] Incorrect Security Classification of Reversible Hardware Modifications
Open 💬 0 comments Opened Jun 9, 2026 by gonzodamus
Bug Description
Personal project classified as cybersecurity threat. Forced off of Fable 5 and back to Opus 4.8. This is a hardware firmware mod that's reversible on a device that I personally own. Project can be found at https://github.com/gonzodamus/FBV_Chroma.
Environment Info
- Platform: darwin
- Terminal: Apple_Terminal
- Version: 2.1.170
- Feedback ID: 97664010-9474-4d21-a613-2f388bcd2af3
Summary from Claude
The flagging looks like topic/vocabulary pattern-matching, not an intent judgment. The project legitimately uses terms that overlap with offensive security — binary patching, "code caves," "detours," and recomputing a checksum so the boot integrity check passes — but these are standard hardware reverse-engineering concepts. The switch notice even admits it "may flag safe, normal content," which is what happened. The switch is content-triggered, so no phrasing or /config setting keeps the work on Fable — it overrides the user's selected model on every message about the project. The effect is that an entire category of legal, non-malicious work (firmware modding of personally owned hardware) is unavailable on Fable. Request: refine the classifier to reduce false positives on benign hardware/firmware modding, or allow user-supplied context ("personal mod, reversible, my own device") to factor into the routing.