[Bug] Overly broad safeguard filter on Claude 3.5 Sonnet blocking legitimate software development tasks
Status Open
Reported on v2.1.211
Maintainer reply None cached
Activity 0 comments · opened Jul 19, 2026
Bug Description
False positive on Fable 5 safeguards. I was doing routine software development work — implementing a "disclosure gating" feature for a character-message web app (controlling which persona facts an LLM can reference based on conversation progress). The conversation involves a dating-service reply-drafting tool, which likely triggered the broad filter, but the actual work is standard TypeScript/testing/git workflow with no harmful content. The safeguard flagged a benign message mid-task and forced a switch to Opus 4.8. Please narrow the filter for legitimate coding tasks.
Environment Info
- Platform: linux
- Terminal: wsl-Ubuntu-24.04
- Version: 2.1.211
- Feedback ID: 35d442c6-d12e-44b3-85dc-0e4dcfc95cb8
Errors
[]