Anthropic Makes Auto Mode the Default in Claude Code to Balance Speed and Safety

Anthropic announced that auto mode is now the default setting in Claude Code. This feature replaces manual approval for every command with a classifier that evaluates actions and blocks those that appear potentially harmful, addressing the tradeoff between speed and safety in agentic coding. Internal evaluations indicate the classifier catches more dangerous actions than manual reviews and withstands third-party red-teaming. Consequently, Claude Code sessions can run 9x longer between interruptions. The announcement highlights production use cases: Nuro uses auto mode for long-running autonomous agents and overnight research tasks; Gusto reports that roughly 10% of session transcripts since mid-May 2026 included an auto mode denial, validating its effectiveness; and Garner Health deployed Claude Code to all 550 employees, enabling a standardized software development lifecycle with 'antagonistic research' stages that were previously impractical due to permission overhead.