AI Weekly Malaysia

Back to items Summaries

Auto mode is now the default in Claude Code for Pro, Max, and Team plans

ID
12346
Status
summarized
Published
09 Aug 2026, 6:36 AM
Fetched
09 Aug 2026, 7:21 AM
Provider
Simon Willison
Category
developer-ai
Original URL
https://simonwillison.net/2026/Aug/8/auto-mode/
Source URL
https://simonwillison.net/atom/everything/

Summary

Score
7.5
Created
09 Aug 2026, 7:21 AM
Tags
Audience
developersvibe_codersai_agent_users

What happened

Anthropic is making auto mode the default for new Claude Code sessions on Pro, Max, and Team plans starting August 14th, backed by evals showing auto mode blocks 89% of dangerous actions versus only 13.6% of human reviewers refusing them. A third-party Trajectory Labs evaluation found zero successful attacks across 720 indirect prompt injection attempts against Claude Fable 5, Opus 5, and Sonnet 5 running auto mode, though 11% of harmful actions would still not be blocked.

Why it matters

If you use Claude Code on a paid plan, your workflow changes on August 14th: new sessions will run in auto mode by default, meaning fewer manual approval prompts. Review the eval methodology before trusting the security claims—11% of dangerous actions still slip through, and the prompt injection results, while strong, cover a specific set of 72 scenarios. Decide whether to keep auto mode on or override it for sensitive repos before that date.

Discussion angle

The 13.6% human refusal rate vs 89% auto mode block rate is a compelling argument, but the 11% gap and the limited 72-scenario injection test mean auto mode is better-than-human, not safe—discuss what guardrails you'd still want around production databases and sensitive file systems before letting agents run unsupervised.

Top