Event ·Multi-source

Claude Code 默认权限8月14日起调整为自动模式

Anthropic宣布8月14日起，Pro、Max和Team用户的Claude Code默认权限改为自动模式。该模式用独立分类器实时评估工具调用和Shell命令，拦截破坏性操作。在1053名付费测试者中，自动模式识别出89%的危险命令，而人工审批仅识别13.6%。提示词注入评估中，720次攻击下Claude Fable 5、Opus 5和Sonnet 5全部拦截，GPT-5.6 Sol在Codex自动审阅下成功率为5.83%。

Current read

Anthropic把Claude Code默认权限改成自动模式，危险命令识别率从13.6%提到89%，还防提示词注入。

11 sources68° AI heatLast updated 2026/8/8 01:17:30

Evidence

11 sources

Primary sourceIT之家

2026/8/8 01:17:30View source

Related source 1Simon Willison’s Weblog

2026/8/8 22:36:03View source

Related source 2Decoder

2026/8/8 14:58:57View source

Related source 3Claude Code: GitHub Releases

2026/8/7 04:00:59View source

Related source 4techcrunch

2026/8/9 19:20:32View source

Related source 5shao__meng

2026/8/7 02:27:38View source

Related source 6The Rundown AI

2026/8/7 10:30:32View source

Related source 7Jerry Liu

2026/8/9 02:24:09View source

Related source 8歸藏(guizang.ai)

2026/8/9 02:49:45View source

Related source 9Thomas Wolf

2026/8/6 12:52:34View source

Related source 10arXiv: Anthropic

2026/8/7 08:32:03View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story