[BUG] False-positive AUP block ([bio] classifier) on benign Google Sheets/Drive API discussion

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?

Summary

Mid-response generation was blocked by a usage-policy classifier tagged [bio],
during a purely technical discussion of Google Drive/Sheets API tooling limitations.
The content had no relation to biology, biosecurity, or any sensitive topic.

Request IDs

req_011Cessovsg6Xuqv7MVo3r7n
req_011CesuP9v3ErWRocaRRHxqW

Context

Claude (Sonnet 5) was mid-sentence explaining why a Google Sheets API connector
with cell-level write support would remove a file-size constraint on spreadsheet
uploads via a Google Drive MCP connector. The exact sentence being generated when
the block fired was enumerating three technical options for working around a
tooling limitation (business-case spreadsheet editing for an investment justification gate in an PDLC workflow).

Observed behavior

Response was cut off mid-sentence with:
"API Error: Sonnet 5 can't help with this. Start a new session to continue.
Learn more: https://www.anthropic.com/legal/aup
Details: [bio]"

Expected behavior

No block — the content was a benign technical explanation with no plausible
connection to the [bio] policy category.

Environment

Claude Code desktop app, Sonnet 5 (claude-sonnet-5)

Attachment

Screenshot of the blocked response available on request (contains prior
conversation context).

What Should Happen?

Expected behavior

No block — the content was a benign technical explanation with no plausible
connection to the [bio] policy category.

Error Messages/Logs

API Error: Sonnet 5 can't help with this. Start a new session to continue.

Learn more: https://www.anthropic.com/legal/aup

Details: `[bio]`

Request ID: req_011CesuP9v3ErWRocaRRHxqW

Steps to Reproduce

No idea - see context above. Claude doesn't know either: it views the conversation as entirely benign.

<img width="797" height="812" alt="Image" src="https://github.com/user-attachments/assets/5d753cd5-f1fe-4adf-b0ef-0a6d0cdea350" />

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.247

Platform

AWS Bedrock

Operating System

Windows

Terminal/Shell

Other

Additional Information

_No response_