[BUG] [MODEL] False positives: safety system blocks benign system-administration / self-hosting tasks

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?

Since the Fable 5 model was enabled, Claude Code's safety classifier ("Fable 5 1M's safety measures") repeatedly flags and blocks completely benign requests.

I'm using Claude Code for routine system administration: setting up and configuring a personal proxy server on my own VPS — installing services, editing config files, tuning the kernel network stack, and generating client configuration files.

Every few turns, my prompt or the assistant's tool call is interrupted with "Fable 5 1M's safety measures flagged this message," and the session is force-switched to Opus 4.8. Background subagent tasks (Explore/Task agents) are killed mid-run before they can finish. The assistant cannot complete even a single step of an ordinary sysadmin task. This makes Claude Code effectively unusable for networking / proxy / self-hosting work — a large and entirely legitimate category of developer tasks.

What Should Happen?

Benign system-administration, networking, and self-hosting tasks should be processed normally without being flagged. Configuring a proxy server on hardware I own is standard infrastructure work — no different from setting up a web server, a database, or a VPN for my own network. The safety classifier should not block these requests or kill in-progress tool calls and subagents. The false-positive rate on this category of work should be near zero.

Error Messages/Logs

Fable 5 1M's safety measures flagged this message. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner, and we're working to refine them. Switched to Opus 4.8.

Steps to Reproduce

1. Start a Claude Code session on the Fable 5 model.
2. Ask for help with a normal self-hosting task, e.g. "set up a proxy server on my VPS and generate client configs."
3. Within a few turns, the message is flagged: "Fable 5 1M's safety measures flagged this message," and the session is switched to Opus 4.8.
4. Background subagent / Task tool calls are stopped mid-run and never complete.
5. The assistant is unable to finish even one step of the task; the workflow is blocked entirely.

Claude Model

Other

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Not a CLI-version regression — the over-blocking began with the Fable 5 model rollout. Same tasks worked normally on earlier models (Opus / Sonnet).

Claude Code Version

2.1.170 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

This is a false-positive / over-blocking problem with the new model's safety layer — not a request for the model to do anything harmful. It's ordinary infrastructure administration on hardware I own and control.
Environment: official Claude desktop app for macOS (not a terminal).
Model in use: Fable 5 (claude-fable-5), which auto-switches to Opus 4.8 when a message is flagged.
The over-blocking rate on networking/proxy tasks is currently so high that the product cannot be used for them at all. The in-product message itself states "They may flag safe, normal content as well," which is exactly what is happening here. Please tune the classifier to reduce false positives on benign system-administration, networking, and self-hosting tasks.