Only within authority your team has granted. Observe mode is on by default, so agents analyze and record verdicts without taking action at first. After you verify performance, you can configure what each agent may do, action by action. Consequential actions go to your team for approval, and the agent carries out what you authorize.

Observe mode lets the agent run on your real environment and show every verdict and the action it would have taken, without taking that action. It’s a verification step, not the end state: once your team is confident in the agent’s calls, you can authorize specific actions.

Nudge Security AI Agents work from context most tools don’t have: discovery across the browser, inbox, identity provider, and connected apps. They reach risk surfaces others miss, like OAuth grants and browser extensions, and they do more than produce a score. They return a verdict and, when authorized, carry out the next step.

Depending on the agent and the authority you configure, Nudge Security AI Agents can revoke OAuth grants, disable browser extensions, permit or resolve what’s safe, send nudges to employees for justification or added context, and record the outcome for auditability.

A person authorizes every consequential action. Clearly safe verdicts can resolve without human review, which is how agents reduce the queue instead of simply adding another layer of analysis.

Start in observe mode, review the agent’s verdicts against what your team would have decided, and then authorize only the actions you’re comfortable with. Every verdict includes reasoning, evidence gaps, and an audit trail so teams can verify how the agent reached its call.