From August 14, Claude Code will use auto mode as the default permission mode for new sessions on Pro, Max, and Team plans, meaning fewer approval prompts and a classifier guarding risky commands instead.
What Changes
Starting August 14, Claude Code will stop pausing for approval on every command for most paid users. Auto mode becomes the default permission mode for new sessions on Pro, Max, and Team plans, according to Anthropic’s official documentation.
Auto mode lets Claude Code run without routine permission prompts by routing tool calls through a classifier that blocks anything irreversible, destructive, or aimed outside your environment. Deny rules and explicit-ask rules are evaluated before the classifier and still block or prompt, so the hard guardrails survive the change.
The classifier is the key difference. Anthropic reports that in testing, the auto-mode classifier caught 89% of dangerous commands, while manual approval caught only 14%. That gap is the whole argument for the change: asking the human every time is not actually safer, because humans rubber-stamp most prompts.
What You Can Configure
Organizations can tell the classifier which repositories, buckets, and domains they trust through the autoMode settings block, so routine internal operations stop getting blocked. There are CLI subcommands to inspect the effective config, review denials, and override block and allow rules. Enterprise setups can also enforce settings through server-managed configuration.
If you rely on explicit approval for sensitive operations, the change does not remove it: deny and explicit-ask rules still take priority over the classifier, and auto mode remains a permission mode you can switch out of per session.
What This Means for Your Workflow
For everyday sessions, the visible difference is speed: a command that used to stall on an approval prompt now runs, with the classifier deciding whether it is safe. That changes the feel of long refactoring sessions and batch operations, where the approvals were noise rather than safety.
For teams, the default flip matters at onboarding time. New users inherit auto mode instead of having to opt in, so the settings that protect sensitive work, like deny rules and explicit-ask rules, become the deliberate layer rather than the default one. Admins can still lock down behavior with server-managed settings if the standard default is too permissive for their environment.
The trade-off is worth stating plainly: the safety checkpoint moves from a human who is often on autopilot to a model that is always paying attention. Anthropic’s own testing, 89% of dangerous commands caught by the classifier versus 14% by manual approval, is the evidence for the swap.
Why It Matters
This is the clearest signal yet that Anthropic believes the classifier is better at risk judgment than the average prompt fatigue. For teams, it means Claude Code sessions will feel faster, but it also shifts the safety burden from a human checkpoint to a model-based one. If you are budgeting for the change, our breakdown of Claude Code token pricing explains what the Pro, Max, and Team tiers actually cost in practice.