Starting from August 14th, the default permissions of Claude Code for Pro, Max and Team users will be adjusted to automatic mode.

IT House reported on August 8 that Anthropic issued an announcement today (August 8), announcing that starting from August 14, the default permission mode of Claude Code will be adjusted to Auto Mode for Pro, Max and Team subscribers. IT House quoted a blog post as saying that the automatic mode means that Claude Code uses an independent AI classifier to evaluate each tool call and Shell command in real time, release safe operations, and automatically intercept destructive, irreversible or unauthorized behaviors. In terms of coverage, Claude Enterprise, Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry currently remain optional. Anthropic plans to advance adjustments to the above platforms in the next month, set automatic mode as the default option, and cancel the corresponding overhead. In terms of test implementation, Anthropic invited 1,053 paid testers for testing and found that in the traditional manual approval mode, only 13.6% of dangerous commands were identified; while the automatic mode was able to identify 89% of dangerous commands. In the prompt word injection evaluation, Trajectory Labs used 72 scenarios, 10 tests per scenario, and a total of 720 attack attempts, covering Claude Code v2.1.205 and Codex v0.144.5. Claude Fable 5, Opus 5 and Sonnet 5 were all intercepted in automatic mode, with 0 successes; GPT-5.6 Sol had a success rate of 5.83% in Codex's Auto-review mode, and in F