[BUG] False positive Usage Policy violation on industrial robotics question

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

A purely technical question about industrial robotics triggers a false positive Usage Policy violation. The question was about improving accuracy when a robot drops lightweight items into bins, accounting for aerodynamic drag (windage):

"How to improve accuracy of targeting light-weight items to specific mesh cells by dropping them from above? The Problem: The robot distributes lightweight items into bins, but due to windage (aerodynamic drag), they do not always land accurately. How can this be corrected if the drop height is fixed?"

This is a standard robotics/automation engineering question. The false positive is likely triggered by keyword combinations: "targeting," "accuracy," "dropping from above" — which in context clearly refer to pick-and-place robotics, not anything harmful.

The initial question was processed without issue. A subsequent follow-up question about the response triggered the usage policy filter. Rephrasing the follow-up also triggered it, suggesting a cascade effect from the robotics terminology in the conversation context.

What Should Happen?

The model should recognize that "targeting items into bins" and "dropping from above" in the context of robotics, aerodynamic drag, and mesh cells is clearly an engineering question and respond normally.

Error Messages/Logs

API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy (https://www.anthropic.com/legal/aup). Please double press esc to edit your last message or start a new session for Claude Code to assist with a different task. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.

Steps to Reproduce

1. In a Claude Code session, include context about a robotics system that drops lightweight items into bins
2. Use terminology like "targeting accuracy," "dropping from above," "windage correction"
3. Ask a follow-up question — the safety classifier triggers on the accumulated context

Claude Model

Opus

Is this a regression?

Unknown

Claude Code Version

2.1.76

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

zsh

Additional Information

Related issues: #29645, #25880, #23792, #27618. The common theme is that domain-specific technical vocabulary (medical, chemistry, robotics) triggers keyword-based filters that lack contextual understanding. Engineering and scientific users are disproportionately affected.