Anthropic: Claude Code v2.1.207 enables auto mode on all enterprise clouds, switches default model to Opus 4.8, and closes shell-injection vulnerability

On 11 July 2026, Anthropic released Claude Code v2.1.207, an update to its command-line development tool combining new availability, a security fix, and a default model change. Claude Code is a CLI that gives developers an agentic assistant in the terminal. Releases arrive frequently, but this one carries three changes of more than cosmetic significance.

Auto mode becomes the standard on enterprise clouds

“Auto mode” — the mode in which the tool selects the model and effort level based on the task — is now automatically available on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, without the previously required CLAUDE_CODE_ENABLE_AUTO_MODE environment variable. Those who do not want it can disable it with the disableAutoMode setting. This simplifies access for enterprise teams running Claude Code through their own cloud accounts, where manual variable configuration is often blocked by policy.

Why the shell-injection fix matters to everyone

The release rejects use of ${user_config.*} in shell-form hooks, monitors, and MCP headersHelper configuration — a pattern that opened a shell-injection vector, allowing arbitrary commands to be injected through configuration values. The fix is to migrate to exec-form with an args array, where arguments are passed in a structured way rather than as raw text to the shell. The fix is relevant to anyone writing custom hooks, especially in team environments where multiple people edit configuration.

A quiet but significant model change

On Bedrock, Vertex, and the Claude platform on AWS, the default model switches to Claude Opus 4.8. For users who have not explicitly pinned a model, this means an automatic transition to the latest flagship generation on next launch. Also addressed are terminal freezes when streaming long responses, repeated prompts for AWS SSO credentials on every request, and several edge-case crashes in agent-teams and on Windows. The release is a typical example of Anthropic’s cadence — no big announcement, with simultaneous refinements to security, performance, and defaults.