Claude Code 2.1.175 and 2.1.176: Enterprise Model Enforcement and a Wave of Remote Control Fixes

Anthropic pushed two Claude Code releases on June 12. The first, 2.1.175, is a single-feature drop aimed at enterprise administrators. The second, 2.1.176, is one of the larger bug-fix releases in recent memory — over twenty fixes, with Remote Control taking the most attention.

2.1.175: Managed Model Enforcement

The new enforceAvailableModels managed setting does two things:

1. When enabled, the Default model is constrained to the availableModels allowlist. If the Default would normally resolve to a model outside that list, Claude Code falls back to the first allowed model instead.
2. User or project-level settings can no longer widen a managed availableModels list. Previously, an administrator could set an allowlist, but individual users could override it from their own settings. That escape hatch is now closed when enforceAvailableModels is on.

This is for organizations that need to control which models their developers use — for cost management, compliance, or to limit access to specific model tiers. Before this change, availableModels constrained what appeared in the model picker, but it didn’t prevent the Default model from resolving to something outside the list.

2.1.176: Closing the Loophole, Plus Much More

Version 2.1.176 extends the model governance work and adds a raft of other fixes.

Model enforcement: The release closes a bypass that existed for environment variables. With 2.1.175, a user could still redirect the Default model to a blocked model by setting ANTHROPIC_DEFAULT_*_MODEL environment variables — a way to override the allowlist without touching user settings. Version 2.1.176 blocks that path. /fast now also refuses to toggle when it would switch to a model outside the allowlist.

Fable 5 auto mode: Organizations that have enabled Fable 5 but don’t have Opus 4.8 available were hitting failures in auto mode. The classifier now falls back to the best available Opus model instead.

Session language: Session titles are now generated in the language of your conversation. Previously, titles were always English regardless of what language you were working in. You can pin a specific language via the language setting if you want to override the automatic detection.

Remote Control: This release fixes over ten Remote Control bugs. The main issues resolved:

• Connecting from web or mobile was silently switching the session’s model to a different one
• Disconnect notifications were showing a numeric code instead of a human-readable reason
• Connection failures were adding a duplicate line to the conversation transcript
• Sessions weren’t disconnecting properly when you signed in to a different account

tmux and SSH: Developers working in tmux over SSH can now copy from /copy and mouse selection reliably — the fix reaches the system clipboard. The tmux paste buffer load is also fixed for versions older than 3.2.

Background sessions: Several background session bugs were resolved, including sessions showing “Working” forever when /bg had nothing left to continue, and a search bug where PRs opened during scheduled wakeups weren’t appearing in claude agents search.

Linux sandbox: A bug where the sandbox failed to start when .claude/settings.json was a symlink with an absolute target path is now fixed.

Hook path matching: The documented patterns for hook if conditions (like Edit(src/**), Read(~/.ssh/**), Read(.env)) now match correctly. They weren’t working as documented before this fix.

A Note on 2.1.177

A third release, 2.1.177, landed June 13 with a changelog and feed update. Based on the commit message, it’s a housekeeping release with no new features.

Sources

• Claude Code Changelog
• GitHub Releases — anthropics/claude-code