Auto mode is now Claude Code's default: what the classifier approves, and how to switch back
Starting today — August 14, 2026 — auto mode is the default permission mode for new Claude Code sessions on Pro, Max, and Team plans. If you've never touched your permission settings, your next session starts with a one-time switch prompt, and after that Claude runs most actions without asking you first. That's a real behavior change, not a UI tweak. This post covers what the classifier actually…
Starting on August 14, 2026, Claude Code sessions for Pro, Max, and Team plans will automatically begin in auto mode. This means that Claude will run most actions without requiring explicit human approval. Users will see a one-time switch prompt after their first session, and from that point forward, Claude will operate automatically.
This is a significant change from previous behavior. Auto mode routes actions through a separate classifier model that reviews each action before it runs. The classifier blocks actions that escalate beyond your request, target infrastructure you don't recognize as yours, or contain hostile content. It still operates alongside two other permission rules: permissions.deny rules block outright, and permissions.ask rules force a prompt.
Auto mode changes the default for actions that don't have explicit rules. For example, pushes and PRs are now allowed by default, even though they weren't before. However, certain types of actions, like force pushes, commits with secrets, or history rewrites, are still blocked. Users can add content-scoped ask rules to keep human checkpoints where they want them.
To switch back to manual mode, users can press Shift+Tab or set permissions.defaultMode to manual in their settings. Organization administrators can also disable auto mode for all users by setting permissions.disableAutoMode to disable in managed settings.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- Launch of DeepSeek’s Harness marks its strategic pivot towards autonomous agentic AI scmp.com
- DeepSeek open sources an agent harness where everything is a plugin thenewstack.io
- DeepSeek Harness developer preview deepseek.com
- DeepSeek releases DeepSeek Harness under the MIT license in developer preview, touting a design where "every capability is a plugin" that can be swapped out (Carl Franzen/VentureBeat) venturebeat.com
- Anthropic details multiagent experiments showing Claude agents can wage a "turf war" over incompatible goals, fail to coordinate, collude on prices, and more (Rebecca Bellan/TechCrunch) techcrunch.com