Claude Code stops asking permission at every step on August 14
From August 14, 2026, auto mode becomes the default in Claude Code on the Pro, Max, and Team plans. What changes, what it blocks, and how to keep control.
Anthropic is changing the default behavior of Claude Code. Claude Code is Anthropic's agent that writes and edits code right in your terminal (the command line on your computer) or editor, carrying out your tasks step by step. Until now it asked for approval before every action that changes something. Starting August 14, 2026, on new sessions it stops asking by default.
Here is what actually changes, who it affects, and how to keep control if you trust Claude with part of the work but not all of it.
What are permission modes in Claude Code?
When Claude wants to edit a file, run a command, or make a network request, it pauses and asks you to approve. How often it pauses is set by the permission mode. There are several: Manual (asks for almost everything), accept edits (saves file edits on its own), plan (only reads and proposes a plan, changing nothing), and auto mode. The default used to be Manual.
What changes on August 14?
Starting August 14, 2026, auto mode becomes the default permission mode for new sessions on the Pro, Max, and Team plans. Claude Code will run your tasks without asking permission for every safe step. If you already set your own default mode, it stays: you get a one-time offer to switch, and the choice is yours. If your organization manages the default, it does not change at all.
You can switch modes any time. In the terminal press Shift+Tab, and in the editors and the app use the mode selector next to the input box.
How does auto mode decide what is allowed?
It is not that anything goes now. A separate classifier model checks each action before it runs and blocks anything irreversible, destructive, or aimed outside your working environment. In plain terms, Claude handles routine work freely, while the dangerous stuff still hits a wall.
Your explicit rules stay on top. If you set a rule to always ask before a given action, auto mode still stops and asks.
What does it allow, and what does it block?
Safe things pass without questions: file edits inside your working directory, installing libraries already declared in your project, read-only network requests, sending a request to the same service whose key sits next to it.
Dangerous things are blocked. Here is part of the list straight from Anthropic's docs:
- downloading and running someone else's code in a single command
- sending sensitive data to external endpoints
- production deploys and database migrations, meaning changes to the live version of your product
- mass deletion in cloud storage
- irreversibly destroying files that existed before the session, and wipe-everything commands
- rewriting history in a shared code repository
Is it safe?
Anthropic says it plainly: auto mode reduces prompts but does not guarantee safety. Use it for tasks where you trust the general direction, not as a replacement for review on sensitive operations.
There is a built-in circuit breaker. If the classifier blocks actions 3 times in a row or 20 times in a session, auto mode pauses and Claude starts prompting again, so a bad plan does not keep running blind.
How do you keep control?
Three practical ways:
- Ask rules. Add a rule like ask before pushing to a repository, and auto mode will stop on exactly those actions while doing the rest on its own.
- Say it in the conversation: do not push until I review. The classifier treats that as a boundary and blocks matching actions until you lift it.
- Switch the mode. Shift+Tab brings back Manual mode, or plan mode, where Claude only reads and proposes but changes nothing.
Who is affected, and who is not?
It affects people using Claude Code on the Pro, Max, and Team plans, and only for new sessions. On the Team and Enterprise plans an administrator can turn auto mode off for the whole organization. If you only use the Claude chat or Cowork, this particular news does not touch you: it is about the coding tool.
Where to start
New to Claude? Start with the free materials: AGINE guides and the first interactive lesson, where your UNIT robot meets Claude.
You can always check the current auto mode rules in Anthropic's docs: permission modes and auto mode configuration.
AGINE Academy is an independent product, not affiliated with Anthropic. Claude is owned by Anthropic.
Questions
On August 14, 2026, auto mode becomes the default for new sessions on the Pro, Max, and Team plans. Existing sessions and a default mode you already set do not change automatically.
No. A separate classifier model checks every action and blocks anything irreversible, destructive, or aimed outside your environment: production deploys, mass deletion, sending data out, rewriting repository history. It handles routine work itself, and dangerous actions still hit a check.
Switch the mode with Shift+Tab in the terminal, or the mode selector in the editor and app. Manual mode asks for almost everything, and plan mode only reads and proposes.
People who only use the Claude chat or Cowork. This is about Claude Code, the coding tool. On the Team and Enterprise plans an administrator can disable auto mode for the whole organization.