anthropics/claude-code v2.1.187

TL;DR

• Security hardening: Anthropic's Claude Code now blocks sandboxed commands from accessing credential files and secret environment variables through a new sandbox.credentials setting
• Enterprise controls: Organizations can now restrict which AI models developers can use within Claude Code, with enforcement across the CLI and UI
• Reliability improvements: Critical fixes address remote tool timeouts, structured output loops, and session resumption bugs affecting production workflows

What happened

Anthropic shipped Claude Code v2.1.187, a maintenance release focusing on security, enterprise governance, and stability improvements across its AI-assisted coding environment. The update arrives as enterprises increasingly integrate Claude into development workflows and require stronger controls over model access and data handling.

The most significant addition is a sandbox.credentials configuration option that prevents sandboxed code execution from reading sensitive files like .ssh, .aws, and environment variables containing secrets. This addresses a critical gap in credential isolation—previously, code running in Claude Code's execution sandbox could potentially access authentication material.

On the enterprise side, organizations can now enforce model restrictions organization-wide. When restricted models are selected via the model picker, CLI flag (--model), or environment variable (ANTHROPIC_MODEL), users see a "restricted by your organization's settings" message, giving IT teams centralized control over which Claude variants developers can access.

The release also fixes several reliability issues: the --resume flag no longer fails when prior runs produced no model interactions; structured output no longer loops indefinitely after successful schema calls; and remote Model Context Protocol (MCP) tool calls that previously hung for five minutes now properly abort with errors (configurable via CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT).

These changes suggest Anthropic is prioritizing Claude Code's readiness for enterprise adoption, where security isolation, compliance controls, and production reliability are non-negotiable requirements.

What happens next

Developers should review the sandbox.credentials documentation if deploying Claude Code in shared or untrusted environments. Enterprise customers should check their organization's model configuration to understand which Claude variants their teams can access going forward. This article does not contain affiliate links.