← All changelogs v2.1.173 claude · claude-sonnet-4-6
Claude Code · Source-level changelog

Version 2.1.173

This release improves the clarity of safety-refusal messages by making them context-aware: requests blocked for cybersecurity or biology topics continue to name those topics, while all other blocks now show a calmer, less alarming message. It also adds detection support for a new "Fable" model and explicitly disables sandboxing on Windows.

Package @anthropic-ai/claude-code Diff v2.1.172 → v2.1.173Provider claudeModel claude-sonnet-4-6
2
Features & Changes
1
Bug Fixes
0
In Development
0
Env Vars / Flags

Improvements

2 items

Context-aware safety refusal messages #

New
What

When a model declines a request due to safety measures, the error message is now tailored to the actual reason rather than defaulting to a generic cyber/bio warning for every refusal.

Details
  • Refusals specifically triggered by cybersecurity or biology content still display the existing message naming those topic areas.
  • All other refusals now show: "This model has measures that flagged something in this session. This sometimes happens with safe, normal conversations." followed by a note that these measures help bring Mythos-level capability in other areas.
  • The model-switching notification (shown when Claude auto-falls back to a different model) also uses this same category check, so users who triggered a fallback for non-cyber/bio reasons no longer see an unexplained reference to cybersecurity or biology topics.
  • Refusal telemetry now records non-cyber/non-bio categories as "other" rather than passing through raw category strings, which normalizes internal analytics.
Evidence

New generic refusal string (search for "This model has measures that flagged something in this session"); category classifier introduced (search for H === "cyber" || H === "bio"); updated fallback-switch messages reference the same classifier

Fable model support #

New
What

Added detection for a model whose name contains "fable" (case-insensitive), with adjusted response-suffix handling when that model is active.

Details
  • A new helper function identifies when the active model is "fable" by checking H.toLowerCase().includes("fable").
  • When a fable model is in use, the [1m] model-suffix token is stripped from responses rather than appended, matching how the model reports itself to the API.
  • This is internal plumbing that enables correct behavior for a new model; no user-facing command or flag change is required.
Evidence

New fable detector (search for "fable"); response-suffix logic updated (the g7 model-name formatter gains a new early-return branch for fable+[1m])

Bug Fixes

1 item
  • Sandboxing is now correctly reported as unsupported on Windows. The platform support check for the sandbox subsystem returns false immediately when running on Windows, preventing any attempt to initialise or check sandboxing on a platform where it does not work. (search for if (s$() === "windows") return !1 inside the sandbox platform check)