BreakthroughsWednesday, March 25, 2026· 2 min read

Anthropic’s Claude Code gains ‘Auto Mode’ to boost safety and control

Source: The Verge AI

TL;DR

Anthropic has launched an "auto mode" for Claude Code that intercepts and blocks potentially risky autonomous actions. The feature helps developers and users get the productivity benefits of AI agents while reducing chances of data leaks, unintended deletions, or execution of malicious code.

Key Takeaways

  • 1Auto mode for Claude Code flags and blocks risky actions before they run, improving safety for autonomous agents.
  • 2The new setting strikes a balance between constant human oversight and full autonomy, offering safer delegation.
  • 3It reduces real-world risks like file deletion, data exfiltration, or execution of harmful instructions.
  • 4Now available in production, the feature helps developers adopt AI agents with greater confidence.

Safer agent autonomy without losing productivity

Anthropic has rolled out a new "auto mode" for Claude Code, its tool that lets AI act with permissions on users' behalf. Rather than forcing people to either micromanage every step or grant unchecked powers to an agent, auto mode provides an intermediate safety layer that watches for problematic behaviors and intervenes before risky actions execute.

The practical benefits are clear: auto mode inspects planned steps, flags suspicious or dangerous operations — like deleting files, exfiltrating sensitive data, or running untrusted code — and can block them or request confirmation. This reduces the likelihood of accidental damage or vectors for malicious prompt injection, while still enabling agents to perform useful, time-saving tasks.

For developers and teams, the feature makes integrating autonomous workflows less fraught. By lowering the operational risk of granting permissions-level capabilities to agents, organizations can experiment and deploy agent-driven automations more confidently, accelerating productivity gains without compromising safety.

Auto mode represents an incremental but meaningful advance in agent safety: it’s a real-world, deployed control that balances autonomy and oversight, helping broaden safe adoption of AI agents in development and everyday workflows.

  • Launches a middle-ground safety option between full autonomy and constant human supervision.
  • Helps prevent accidental or malicious actions like file deletion and data leaks.
  • Supports safer, faster adoption of agent-driven automation for developers and teams.

Get AI Wins in Your Inbox

The best positive AI stories delivered to your inbox. No spam, unsubscribe anytime.