Do all domestic LLM models have issues with tasks being blocked or rejected due to sensitive information? I’ve been using Zhipu GLM on Lobster and often encounter the content safety filter triggering a sensitive block: Provider finish_reason: sensitive; 400 status code (no body) Summary: Solutions for sensitive error Root Cause The content safety system of the GLM API (Zhipu) returns finish_reason: sensitive, meaning the request or generated content triggered a review. Fixes Applied 1. Adjusted fallback order: glm-4.7-flash → gemini-2.5-flash → deepseek-chat 2. Restarted the Gateway If the issue persists, alternative solutions: Option A: Switch the primary model to Gemini (recommended) Option B: Disable mem9 memory injection (if the sensitive content originates from memory) Option C: Contact Zhipu API to whitelist If the business content is legitimate (e.g., cryptocurrency discussions), you can contact Zhipu to request a relaxation of content review.