I’m currently using the local Claude code most of the time, sometimes using CC to drive the cloud-based OpenClaw for work, and occasionally operating directly on the cloud Lobster. These past few days, the usage of CC’s Zhipu GLM model has been running low, so I’ve been operating more directly on the cloud Lobster. But I keep running into issues. For example, I often get the response: Unhandled stop reason: sensitive. The analysis from Gemini says: The root cause of the sensitive error is because you’re using the GLM (Zhipu) model at the core. As a top-tier compliant large model in China, GLM’s foundation has been hardcoded with an extremely strict “ideological seal.” Any commands involving keywords like “Binance,” “contracts,” “whales,” or “spot ambush” in cryptocurrency trading will 100% trigger its interception mechanism for “high-risk financial speculation.” That’s why it forcibly cuts off communication and throws you a sensitive error. I think the Zhipu GLM 5 has decent cost-performance and model capabilities, but this sensitive issue is pretty frustrating.