Claude Codeが監督なしで動作 ― Anthropicは人間より安全だと主張
AnthropicはClaude Codeのauto modeをデフォルトにし、有害なアクションの89%を検出できると主張。人間のレビュアーは14%にとどまるという。これは、常時監督なしで動作する自律型AIエージェントへの転換を示すものだ。
anthropic is removing friction from ai-assisted development by making claude code's auto mode the default setting for paid users on august 14. the feature allows the model to execute coding tasks autonomously, bypassing permission prompts unless an action is flagged as irreversible, destructive, or external. in a study of 1,053 testers, auto mode successfully blocked 89% of harmful actions compared to just 13.6% caught by manual review, suggesting that ai guardrails can be more reliable than human attention spans. this shift signals growing confidence in autonomous agent safety and pressures competitors to move beyond human-in-the-loop workflows.