Anthropic will make auto mode the default setting in Claude Code for Pro, Max and Team users from August 14, allowing the AI coding assistant to take more actions without requiring developers to approve every step.
The company introduced auto mode as a test feature in March as part of its effort to make AI-assisted programming faster and more autonomous.
Under the new default, Claude Code will generally proceed with coding tasks without repeatedly asking for permission. However, Anthropic said it will still pause before actions considered “irreversible, destructive, or aimed outside your environment.”
Anthropic says auto mode can improve safety
Anthropic said testing showed that auto mode could be more effective at identifying potentially harmful actions than relying on users to manually approve requests.
In a study involving 1,053 paid testers, the company said auto mode detected 89% of harmful actions, compared with just 13.6% detected through human review.
The company also said users approve about 97% of permission prompts in Claude Code, raising concerns that repeated approval requests can lead to users automatically accepting them.
Boris Cherny, head of Claude Code, said his team had been using auto mode exclusively for several months.
New safeguards added
Anthropic said it has introduced additional protections alongside the change, including prompt-injection screening and customizable hard-deny rules that can block Claude Code from performing specific actions.
The safeguards are designed to limit the risks associated with giving AI agents greater control over development environments.
Developers will still be able to control actions that could make irreversible changes or affect systems outside their working environment.
What changes from August 14?
- Auto mode becomes the default for Pro, Max and Team users.
- Claude Code will require fewer approval prompts.
- Safeguards will remain for destructive, irreversible and external actions.
- Prompt-injection screening will provide an additional layer of protection.
- Users can create custom deny rules for specific actions.
The move reflects the broader shift in AI coding tools from assistants that wait for instructions to autonomous agents capable of carrying out multi-step programming tasks with less human intervention.
