Claude Code自动模式将默认开启，编程介入更少

Image Credits: Samuel Boivin/NurPhoto / Getty Images

使用 Claude Code 编程很快将只需要更少的人工介入。Anthropic 日前表示，将从 8 月 14 日起把 auto mode 设为 Pro、Max 和 Team 账户的默认模式。

该公司于今年 3 月首次公布了 auto mode 的测试版本，称这是一种平衡速度与控制力的方式。Anthropic 在周五的公告中解释道，当 Claude Code 处于 auto mode 时，除非某个操作被判定为"不可逆的、具有破坏性的、或指向您当前环境之外"，否则它将直接继续执行，而不会弹出需要人工审批的提示。

Anthropic 还表示，在测试中，auto mode 被证明比人工审查更安全——在涉及 1,053 名付费测试者的研究中，auto mode 捕获了 89% 的有害操作，而人工审查仅捕获了 13.6%。（或许原因在于："人工审查容易流于习惯：用户在 Claude Code 中会批准 97% 的权限提示。"）

Claude Code 负责人 Boris Cherny 在 X 上发帖称："团队和我都只使用 Auto mode，已经持续好几个月了。无法想象回到权限提示框的日子！"

该公司还表示，正在增加新的安全功能，如 prompt 注入筛查和可定制的 hard deny 规则，以防止数据泄露等行为。

Transportation TechCrunch Mobility: Zoox prepares for launch and Uber's AV empire Kirsten Korosec 3 hours ago

TechCrunch Mobility: Zoox prepares for launch and Uber's AV empire

Government & Policy Historian Jill Lepore says Silicon Valley misreads science fiction and undermines democracy Anthony Ha 4 hours ago

Historian Jill Lepore says Silicon Valley misreads science fiction and undermines democracy

AI The AI safety test is becoming a safety risk Rebecca Bellan 5 hours ago

The AI safety test is becoming a safety risk

In Brief Anthropic is turning Claude Code's auto mode on by default Anthony Ha 12 seconds ago

Anthropic is turning Claude Code's auto mode on by default

Government & Policy Historian Jill Lepore says Silicon Valley misreads science fiction and undermines democracy Anthony Ha 4 hours ago

Historian Jill Lepore says Silicon Valley misreads science fiction and undermines democracy

AI The AI safety test is becoming a safety risk Rebecca Bellan 5 hours ago

The AI safety test is becoming a safety risk