On March 24, 2026, Anthropic announced a new feature for Claude Code. Called "Auto Mode," it lets the AI judge and act on its own permissions. Moving away from the old model that required user approval for every action, Claude can now assess safety on its own and proceed with tasks automatically.
The Old Dilemma for Dev Tools
Until now, Claude Code has stuck to a conservative permissions policy, requiring user approval for every file write and every bash command. Safe, but inefficient. Kicking off a large task and stepping away was hard, since the AI would keep stopping midway to ask for approval.
Some developers got around this with the "--dangerously-skip-permissions" flag. But as the name suggests, it was a risky choice, opening the door to file deletion, sensitive data leaks, and malicious code execution. Developers were ultimately forced to pick between safety and efficiency, even though neither was a value they wanted to give up.
How Auto Mode Works
Auto Mode resolves this dilemma, and the mechanism behind it is sophisticated. Before each tool call executes, a classifier reviews the action, screening for potentially destructive operations such as mass file deletion, sensitive data exfiltration, or malicious code execution.
Actions judged safe proceed automatically. Actions judged risky get blocked, and Claude tries a different approach. If Claude keeps insisting on an action that keeps getting blocked, it ultimately asks the user for permission. The structure trusts the AI's judgment while still leaving humans as the final safeguard.
Limitations and What's Next
Auto Mode isn't perfect. It can let risky actions through, and it can block safe ones — especially when user intent is ambiguous or the AI lacks enough context about its environment. It also adds a modest overhead in token usage, cost, and latency, the price of having the AI review every action along the way.
Still, this is only the beginning. The feature is launching as a research preview on the Team plan, with plans to gradually expand it to Enterprise and API users. Notably, it works with both Claude Sonnet 4.6 and Opus 4.6.
From Tool to Collaborative Partner
Auto Mode marks a sign of maturity for AI development tools. Early AI tools were either extremely restrictive or extremely risky — either the user controlled everything, or the AI was handed everything.
Now AI can judge risk on its own and exercise an appropriate degree of autonomy. It's part of the AI's evolution from a mere tool into a collaborative partner — not blindly following rules, but starting to develop the ability to understand context and make judgment calls.
For developers, this opens up new possibilities: kicking off a big task and stepping away with peace of mind, because the AI can steer clear of risk on its own while still getting the work done efficiently.
