Claude Code 'auto mode'

Anthropic released Claude Code’s new “auto mode” as a research preview on March 24, 2026, with access for Claude Team users today and Enterprise and API rollouts promised in the coming days. (techcrunch.com) Auto mode runs a safety classifier that reviews each tool call before execution, automatically allowing low‑risk actions and blocking high‑risk operations such as mass file deletion or signs of data exfiltration. (zdnet.com) Anthropic says the feature currently works with the Sonnet 4.6 and Opus 4.6 model aliases and recommends running auto mode in sandboxed or isolated environments to limit blast radius. (tech.yahoo.com) Anthropic’s engineering team reported earlier that sandboxing reduced permission prompts in internal use by about 84%, a capability auto mode is being released alongside to shorten approval friction for long agentic sessions. (anthropic.com) Developers can enable the preview via the CLI (examples reported as running claude --enable-auto-mode) and toggle modes with Shift+Tab in the terminal, while VS Code and desktop users switch it on via the extension or app Settings; organization admins can disable auto mode through managed settings. (sdtimes.com) Anthropic warns auto mode can increase token usage, latency, and per‑call expense for tool invocations, which may raise costs on high‑volume workflows. (blockchair.com) Independent hands‑on comparisons run this month found Claude Code outpacing Google’s Antigravity and OpenAI’s Codex on multi‑step reliability tests, with an XDA developer test concluding only Claude Code was “worth using” for the specific hackathon task. (xda-developers.com)