Anthropic says auto mode will be the default in Claude Code for Pro, Max, Team plans, starting on Aug. 14, claiming it's good enough at catching harmful actions
Anthropic introduced auto mode for permission-level decisions in March, with safeguards aimed at destructive actions such as mass file deletion. Making it the default turns that opt-in capability into the standard Claude Code workflow for its paid individual and team tiers.
The move follows Anthropic’s effort to let Claude apply broader principles rather than fixed rules through its reworked constitution. Default autonomy makes the reliability of those judgment calls a product-setting issue rather than a feature experiment.
First-order effects
Pro, Max, and Team users will encounter fewer permission interruptions in Claude Code after Aug. 14, while Anthropic’s harmful-action detection becomes the default control layer for those plans.
Anthropic assumes more direct responsibility for deciding which Claude Code actions can proceed automatically and which require intervention.
Second-order effects
Team customers will need to assess Claude Code’s automatic permission decisions as part of their own development controls, rather than treating auto mode as an optional productivity setting.
Competing coding-agent vendors face pressure to pair lower-friction autonomous operation with equally legible safeguards against destructive actions.
Third-order effects
If default-on permission automation becomes standard, coding assistants will increasingly compete on operational assurance: the ability to act quickly while making safety controls credible to paid teams.
The boundary between an assistant that proposes code and an agent that carries out workspace actions will narrow, shifting differentiation toward policy enforcement and auditability.
The trend: AI coding tools are moving from user-approved suggestions toward default autonomous execution, with safety controls becoming the central trust mechanism.
We're making this change for two reasons: 1. In our testing, auto mode matched or beat manual permission review on every safety measure we tracked. 2. It makes long-horizon work more viable. Claude runs longer between interruptions, so you can run multi-hour tasks in the
Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users. Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.
automode is much safer than any other permission system out there, especially reviewing them yourself excited to announce we're rolling it out to everyone by default, with no overhead cost for the classifier [image]
One reason we trust it more than manual approval: in a study with 1,053 paid testers, we swapped a permission prompt for a clearly dangerous command (text only, nothing actually ran). Testers caught it 13.6% of the time, and closer to 5% after 50 prompts. Auto mode blocked the [i…
The team and I use Auto mode exclusively, and have been for many months. I couldn't imagine going back to permission prompts! Really excited to get this out to everyone.
You'll get an in-app notification when the default flips. If you've already set a default permission mode, Claude will ask before changing anything. You can still switch modes with Shift+Tab anytime. Admins can pin defaultMode (or disable auto mode entirely) in managed settings
“it caught 89% of dangerous commands” well... what about the other 11%? being better than humans (who usually won't look carefully) doesn't mean it's useful this proves reviewing each command is simply not the right solution instead, we need to secure the environment and set
Honestly, I think this is a good move The interesting part isn't really the 89% vs 14% number. It's that permission prompts are starting to become a bottleneck for agents that are supposed to work for hours on their own If Claude has to stop and ask me for permission every [image…
Anthropic will make Auto Mode the default in Claude Code for Pro, Max, and Team plans starting August 14, citing a classifier that catches 89% of dangerous commands versus 13.6% for humans. Developers may spend less time writing code and more time monitoring AI output. — Sourc…