Anthropic has introduced “auto mode” in Claude Code, a new permissions setting that allows the AI to execute tasks with fewer manual approvals while using a built-in classifier to screen actions for risk.
The feature, released Tuesday as a research preview for Team users, is designed to reduce interruptions in longer workflows where Claude would otherwise require approval for each file change or command. Enterprise and API access is expected in the coming days.
Auto mode sits between two existing approaches: the default, highly restrictive permission system and a flag that bypasses safeguards entirely. Instead of requiring constant user input or removing checks altogether, the new mode evaluates each action before execution. A classifier reviews operations for risks such as large-scale file deletion, data exfiltration, or harmful code execution.
If an action is considered safe, it proceeds without user input. Higher-risk actions are blocked, prompting Claude to adjust its approach. Repeatedly blocked attempts can eventually trigger a direct permission request to the user.
Anthropic positions the feature as a way to support longer, unattended tasks while maintaining guardrails. However, the company noted the system is not risk-free and continues to recommend use in controlled environments. The classifier may still allow some risky actions in ambiguous situations or when context is limited, and it can also block benign operations.
The feature may introduce slight increases in token usage, cost, and latency for tool calls.
Auto mode works with Claude Sonnet 4.6 and Opus 4.6 and can be enabled via command line, desktop settings, or the VS Code extension. It is disabled by default on desktop and can be managed by administrators through organization settings.
The release reflects a shift in how Anthropic is handling permissions in its coding tools, aiming to balance autonomy and oversight as developers rely on AI systems for more complex, multi-step tasks.
This analysis is based on reporting from Claude.
Image courtesy of Claude.
This article was generated with AI assistance and reviewed for accuracy and quality.
About this article: This article was generated with AI assistance and reviewed by our editorial team to ensure it follows our editorial standards for accuracy and independence. We maintain strict fact-checking protocols and cite all sources.
Get the latest AI news delivered to your inbox every morning.
More articles from Tech Deep Dive
Arm has introduced its AGI CPU, a new data center processor with up to 136 cores, marking the company’s first move into selling its own finished silicon rather than licensing chip designs. The...
Google has introduced an upgraded version of its AI Studio coding environment, adding a new “vibe coding” experience powered by the Antigravity agent to generate production-ready applications...
Cursor has released Composer 2, a new in-house coding model integrated into its AI coding platform, alongside a faster default variant called Composer 2 Fast, marking a significant update to its...
Join thousands of weekly readers — the latest AI news, free in your inbox.