[BUG] False-positive Usage Policy (AUP) violation on benign FPGA/hardware engineering waveform analysis

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?

Claude Code blocked a completely benign technical request with the Usage Policy (AUP) violation error before any response could be generated.

The request was an analysis of embedded hardware control system waveform data — an exported chart image comparing a "normal" vs "abnormal" case of a power-range control state machine (setpoint signals, monitored power, and range-status output). This is ordinary hardware/RTL engineering work (signal-chain measurement, range-switching logic) and contains no malicious, harmful, or policy-relevant content.

The identical content is analyzed without any problem in claude.ai, which strongly indicates the Claude Code safety filter is producing a false positive on benign engineering material.

What Should Happen?

Benign hardware/software engineering analysis should not trigger an AUP violation. The model should describe the waveforms and provide the requested analysis. The safety filter should be tuned to reduce false positives on benign technical/engineering content.

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). Try rephrasing the request or attempting a different approach.

Request ID (from the API response, not a Feedback ID): req_011CbJmgV7CVJscr4ZZpNa5f
Feedback ID: none

Steps to Reproduce

1. In a technical engineering session, attach a chart/waveform image (e.g., an exported plot of control signals with legends like "setpoint", "monitor", "range status").
2. Ask the model to analyze the waveforms: describe the legend, compare a "normal" vs "abnormal" case, and hypothesize a root cause.
3. Observe the AUP violation error instead of a normal response.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.145

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

• Model-dependent: the same request is blocked with Opus 4.7 but completes successfully with Sonnet 4.6, and also works without issue in claude.ai. This suggests the false positive is specific to the Opus 4.7 path rather than a general regression across versions.

• Workaround: switching models (e.g., /model claude-sonnet-4-...) avoids the block.

• The /bug command is not available in this environment ("/bug isn't available in this environment."), which is why this is filed via GitHub instead.

• The content was a hardware-engineering waveform chart (FPGA RF power-range control, normal vs abnormal case). No malicious, harmful, or policy-relevant material was involved.