Anthropic to Change Claude Code Default Setting to Auto Mode Next Week

Anthropic will switch Claude Code's default permission mode from manual approval to auto mode starting August 14 for Pro, Max, and Team plan users . Auto mode routes shell commands and file actions through a separate classifier instead of prompting for approval on each step; anyone with a different existing default gets a one-time prompt asking whether to switch.

The change follows Anthropic's own testing showing the classifier caught 89% of dangerous commands versus roughly 14% for manual human review, with teams reportedly shipping about 25% more pull requests once approval friction eased. Separately, Claude Code v2.1.224, released August 7, lets one session message another active session on macOS or Linux via new ListAgents/SendMessage tools, cutting the need to manually copy context between terminal windows . Anthropic also updated biology-related safety guardrails this week, cutting automatic fallbacks by about 85% in testing, while virology and toxicology stay gated.

Anthropic is privately held, so neither change is directly tradable. The relevance is to the broader agentic-coding-tool landscape: lower approval friction and cross-session coordination could accelerate enterprise adoption of AI coding agents, a category also contested by public developer-tooling players and chipmakers such as NVDA supplying the underlying compute.

Related Stocks

• CRM