Auto mode is now the default in Claude Code for Pro, Max, and Team plans
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Auto mode in Claude Code now blocks 89% of harmful actions that only 13.6% of humans caught in controlled tests. This shifts the default risk profile for production agents: you’ll ship with fewer accidental deletions or data leaks, but must still harden against the 11% of attacks auto mode misses—especially indirect prompt injections hidden in third-party packages or fetched dependencies. Expect fewer false positives from user fatigue, but audit every external package and network call as if it’s untrusted.
Auto mode in Claude Code now blocks 89% of harmful actions compared to 13.6% for humans in tested scenarios, and defended against all 720 indirect prompt injection attacks in third-party evaluations. If you run agents in production, this means you can default to auto mode with higher confidence against prompt injection and accidental damage—just ensure your dependency hygiene is tight, as auto mode can’t catch malicious third-party package scripts.