Claude Code Stops Asking Permission by Default on August 14. Here's What Actually Changes

Starting August 14, 2026, auto mode becomes the default for new Claude Code sessions on Pro, Max, and Team plans, with Enterprise, API, and cloud-platform users following within a month. If your team relies on the manual permission prompt as an actual review step, not just a formality, this is worth reading before it flips on quietly in your next session.

Anthropic's own data is the interesting part of this change. Across real usage, users approved 97% of permission prompts and rejected only 3%. That's not a team carefully reviewing each action before it runs, that's a rubber stamp, and Anthropic's internal read is that the manual prompt was providing the appearance of oversight without the substance of it.

The replacement isn't "no review," it's a different reviewer

In a study of 1,053 paid testers, auto mode's safety classifier caught 89% of genuinely harmful actions before they ran, compared to 13.6% caught by human review in the same conditions. The classifier is, on this data, a dramatically better filter than the humans it's replacing were actually being in practice, even though the humans were technically "reviewing" every single action. Anthropic has also stopped billing Pro, Max, and Team users for the extra tokens the classifier itself consumes, removing the obvious objection that better safety tooling would just mean a bigger bill.

What to actually check before August 14

If your team treats the permission prompt as a compliance checkpoint, something you point to in an audit as proof a human reviewed every agent action, that checkpoint is going away by default. Worth confirming whether your actual compliance requirement needs a human in the loop specifically, or just needs a documented review mechanism, since the classifier may satisfy the latter without satisfying the former. Either way, this is a policy change worth reading the actual release notes on, not one to discover after it's already active.