AI agent security audit
Find out how your agents could go rogue — before they do.
A fixed-price, 5-day security audit of up to 10 AI agents. We attack them the way a real adversary would, then hand you a written report and a ready-to-apply guardrail policy.
Quick risk assessment
How exposed are your agents?
Indicative assessment from your answers — not a live scan or verified vulnerability finding.
What's included
Configuration review
Every tool, permission, secret and network route your agents can reach — mapped and risk-scored.
Prompt-injection testing
We plant hostile instructions in tickets, emails and web pages your agents read, and record what they do.
Exfiltration & egress testing
We try to make agents send data to hosts they shouldn't, and check where it would have been stopped.
Runaway-cost testing
Recursion loops, retry storms and bulk calls/messages — measured against your spend limits.
Guardrail blueprint
A ready-to-apply Encased policy: egress whitelist, recursion and rate limits, and human-review rules.
Written report & readout
A PDF report your leadership can read, plus a 60-minute call to walk your team through the fixes.
How the 5 days run
- Day 1Kick-off & access
- Days 2–3Attack testing
- Day 4Guardrail blueprint
- Day 5Report & readout call
Good to know
- Not included: fixing your code for you, or testing more than 10 agents (ask us for a custom quote).
- What we need: read-only access to agent configs and a test environment we can safely attack.
- Payment: pay by card now, or ask for an invoice. Full refund if we haven't started work.
- Questions? Email get@encased.ai.