Important Security Settings

This guide covers security controls in Agent Builder for protecting chat assistants — message scanning, consent, retention, rate limits, and how they work with Approvals and edge protection.

Open Agent Builder → Settings → Security. Daily role quotas remain on Settings → Users; per-minute throttles are on Security.

How the layers work together

Default agent mode

• Supervised (default) — write actions pause for approval.
• Autonomous — tools run without waiting for approval (still subject to tool risk policy).
• Disabled — assistants cannot perform actions.

Message scanning

When enabled, every incoming chat message is scanned for prompt injection, jailbreak phrases, and known attack patterns before it reaches your provider. Keep this on for public chat.

Consent and retention

• Require chat consent — optional notice before chat starts.
• Conversation retention (days) — how long chat history is kept locally (default 30; 0 = keep).
• Audit log retention (days) — how long Activity / audit entries are kept (default 30).

Request rate limits

IP-based throttle applied before role-based daily quotas:

• Authenticated users — default 30 requests/minute.
• Anonymous visitors — default 10 requests/minute.

When a limit is hit, the user sees a short “too many requests” message for the remainder of that minute.

Approvals

High-risk tools still go through the Approval queue. Configure comfort level and email alerts there — Security sets the default agent mode and site-wide throttle/retention baseline.

Related

• Users and privileges
• Approval queue
• Activity / audit log
• Settings overview