[Bug] Anthropic API Error: Overly aggressive cyber-related content filtering on authorized security research

**Bug Description**
req_011Cecvs357nXk7GNMwg2otj I was conducting authorized security research for a HackerOne bug-bounty program, within the program’s scope and rules. Opus 5 flagged my request as cyber-related even though I was not attempting any unauthorized access or harmful activity.

This seems like a false positive. Please improve the distinction between legitimate, authorized vulnerability research—such as code review, HTTP analysis, report writing, and remediation advice—and harmful cyber activity. If possible, please indicate what triggered the flag so I can formulate future requests more clearly.

**Environment Info**
- Platform: darwin
- Terminal: vscode
- Version: 2.1.252
- Feedback ID: 6c6d9d06-76f7-43b5-ad2f-fe7a032d2f4b

**Errors**
```json
[]
```

Start by reviewing the report associated with feedback ID 6c6d9d06-76f7-43b5-ad2f-fe7a032d2f4b and reproducing the flagged request in the reported VSCode-on-darwin environment. The issue provides no source file or test entry point; done would require a clearer distinction between authorized vulnerability research and harmful cyber activity, plus useful information about what triggered the flag.

索引モデルが issue の本文から書いたものです。