When deploying Claude in an agentic context where it can take autonomous actions (send emails, modify files), what safety principle is MOST important?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Agentic AI follows the same rule as computer security: least privilege. Give Claude exactly the access it needs for the task — no more. And prefer actions you can undo if something goes wrong.
Full explanation below image
Full Explanation
In agentic deployments, applying least privilege is critical: give Claude only the permissions actually needed for each specific task (not blanket admin access), prefer actions that can be reversed over irreversible ones (e.g., move to trash instead of permanently delete), maintain human oversight especially for high-impact actions, design explicit check-in points for ambiguous situations, and scope permissions to the minimum necessary. Option A (maximum permissions) violates least privilege and creates huge risk surface. Option C (maximize speed) ignores safety. Option D (unlimited trust post-eval) ignores the fact that deployment context differs from evaluation and errors will occur.