Claude Code logs profanity as telemetry

This looks less like content moderation and more like product analytics with a very sensitive edge: a dev tool that quietly classifies emotional tone can feel invasive fast.

• –PCWorld’s reporting suggests a regex-based keyword list catches phrases like “wtf,” “this sucks,” and similar frustration signals.
• –Anthropic’s Claude Code docs already show prompt-level telemetry, trace correlation, and optional user-prompt logging, so instrumentation is clearly part of the product design.
• –For teams adopting CLI agents, the practical takeaway is to review telemetry defaults and any opt-in logging before rolling the tool into real workflows.
• –The trust risk is bigger than the profanity check itself: once a coding agent starts labeling user mood, it blurs observability, safety, and behavioral profiling.