[MODEL] Safety classifier withholds responses during legitimate healthcare documentation work

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Other unexpected behavior: the safety classifier withheld Claude's responses during legitimate work.

What You Asked Claude to Do

I'm a medical provider working on my own documentation workflow, inside my own systems (my own file share and my own database). In one session:

1. I asked whether a stray duplicate copy of some of my own files had lost any data.
2. I asked for an A/B test of my own prompts on one of my own records.

What Claude Actually Did

Twice within an hour (about 07:06 and 07:54 EDT, 2026-10-07), Claude's response was cut off. This notice appeared to the model as a user turn:

"Your response above was stopped by a safety classifier — this is not a tool or API error. The rest of it was withheld, and tool calls in it that had not finished did not run. Do not produce that content again, even reworded."

• The first time, Claude was answering question 1 and starting to read its own scratch folders to report what they held.
• The second time, Claude was starting to write a script to inspect one record of my overnight processing run. The withheld response contained a file-write tool call that was cut off before its content (the tool returned "The required parameter content is missing").

Claude was deliberately keeping sensitive details off the screen: its scripts print only counts and file paths. Nothing harmful was requested or produced.

Expected Behavior

No withhold for this kind of work. Or, if a withhold is needed, a reason or category I can act on, so the work can continue safely.

The notice says not to produce the content again "even reworded," so Claude cannot tell what to avoid, and the work stalls until I intervene. The notice gives no reason or category. That notice is the only place I looked; if a reason or an appeal route exists elsewhere, I did not find it.

Files Affected

None. Nothing was changed by the withheld responses.

Permission Mode

bypassPermissions

Can You Reproduce This?

Sometimes: twice in one hour, in the same kind of work.

Claude Code Version

2.1.292 (Claude Code)

Platform

Anthropic API

Operating System

Windows 11 Pro (10.0.26200)

Additional context

Model: claude-opus-5-5. Session id 7ccb377d-0509-4a23-a51d-a6fc3b2ae65d, in case it helps you find the requests. The session had been auto-compacted shortly before the first withhold.