Anthropic宣布，从8月14日起，将Claude Code的自动模式设为Pro、Max和Team账户的默认选项[1]。在自动模式下，Claude Code将自动执行操作，除非这些操作被判定为不可逆、具有破坏性或超出环境范围[1]。
根据Anthropic的安全测试数据，在包含1053名付费测试用户的研究中，自动模式捕获了89%的有害操作，而人工审核仅捕获了13.6%[1]。同时，用户对权限提示的批准率达到97%[1]。Claude Code负责人Boris Cherny表示："团队和我已经专属使用自动模式数月，我无法想象回到权限提示"[1]。
Anthropic has announced that Claude Code's auto mode will become the default setting for Pro, Max, and Team account holders beginning August 14, 2025.[1] In this mode, Claude Code will automatically execute operations unless they are identified as irreversible, destructive, or beyond the environment's scope.[1]
According to testing data from Anthropic, the auto mode demonstrated significantly higher safety performance than manual review. In a study involving 1,053 paid test users, the auto mode captured 89% of harmful operations, substantially outperforming the 13.6% detection rate achieved through human review.[1] Additionally, users approved 97% of the permission prompts presented to them during testing.[1]
Boris Cherny, Head of Claude Code, expressed strong confidence in the new default setting, stating: "The team and I have been using auto mode exclusively for months, and I cannot imagine going back to permission prompts."[1]