Anthropic makes Claude Code auto-mode default in 5 days, ending manual permission approvals
量子位 · wechat · 2026-08-10
Anthropic announced that all Claude Code will default to auto-mode in 5 days, replacing manual permission approvals. Key data:
- Users approve 97% of permission prompts, reject only 3%.
- In controlled experiments, humans blocked dangerous commands 13.6% of the time, while auto-mode blocked 89%.
- In production data, harmful actions at severity 7+ occurred in 6.3% of manual sessions vs 2.4% in auto-mode.
- Collaboration with Apollo Research reduced miss rate from 12% to 7%.
- Third-party tests: auto-mode had zero successful attacks in 720 attempts, while GPT-5.6 Sol had 5.83% success in Codex Auto-review mode.
Auto-mode consumes extra tokens per tool call but is free for users. Cloud platforms (AWS, Google, Microsoft) have one month to switch. Users can toggle with Shift+Tab; admins can set defaults.
Related event: Anthropic to Enable Claude Code Auto Mode by Default on August 14(2 posts)→
More from coding & agent
- Developer Teases Upcoming Open-Source Release of App with Persistent Terminal — zhengyiluo · 2026-08-10
- AgentsDock: Open-Source Tool to Orchestrate Claude Code from Your Phone — zhengyiluo · 2026-08-10
- Steadwing Brings Coding Agents On-Call with New Agent Skill and SDKs — khant_dev · 2026-08-10
- AI Coding Agent Blows Through $19 Instantly, Highlighting Cost Control Risks — teortaxesTex · 2026-08-10
- Ex-OpenAI Advisor: Gradual Milestones Are Normalizing AI Self-Improvement — Miles_Brundage · 2026-08-10
- Awesome Agents: A Curated GitHub Directory of Open-Source AI Agent Tools — tom_doerr · 2026-08-10