Claude Max user weighs cancellation over safety-classifier refusals
A Reddit user said Claude repeatedly classified a legitimate OpenWrt porting task for their UniFi U6 Pro router as cyber-related and refused to assist. The user said previous prompt adjustments no longer worked and that the refusals left them without progress on the project.
The report highlights a user-experience problem with Claude's safety classifier on security-adjacent reverse-engineering work, although it is an unverified single-user account.
Just wanted to vent a bit. I do a lot of RE work, and claude has been very useful for that. I got used to the "safety" classifier misfiring on things that were vaguely security-adjacent. Adjusting my prompt or approaching from another direction usually got past it.
This has been getting steadily worse over time, and my patience is almost at an end. Today I was trying to port OpenWRT to my Unifi U6 Pro. Extending the life of my own equipment. And claude just isn't having it. It keeps flagging my sessions cyber and refusing, and none of the usual workarounds are helping. This is a completely legitimate use case, so much so that it's actually explicitly granted by the right-to-repair law in the EU/EEA. I'm so pissed off after wasting hours not making progress on my project.
A tool that refuses to work 50% of the time is not useful. Let alone for $200/month. Fix this shit, claude, or you'll end up hemmorraghing customers.