The deliverables, the findings, the report — and the checklist that scores any audit quote in five questions.
Short answer: a real $6,000 AI security audit hands you four things: a mapped attack surface for your deployed agent, manual prompt-injection and tool-misuse testing against it, a written findings report with reproductions you can hand to your engineers, and a prioritised remediation roadmap. If a quote near that number skips any of the four, it's a scanner run with a logo on it.
This page is the audit I sell, laid out in the open — because the honest way to buy one is to know what lands in your inbox at the end. I also keep a full sample report live on this site, so you can read the actual deliverable before you pay anyone anything.
A $6,000 flat-fee audit is roughly a week of senior human time plus report writing. Here's the deliverable list I work to (all of it visible in the sample report):
| Deliverable | What it contains |
|---|---|
| Executive summary | Severity counts (Critical / High / Medium / Low), a one-paragraph bottom line, and a go/no-go: whether the agent is production-ready for untrusted input |
| Methodology & rules of engagement | Exactly what was tested, what was off-limits, and how much of the surface was covered |
| Attack surface mapping | Every prompt entry point, retrieved-content path, tool schema, and data store the agent can touch |
| Findings with reproductions | Each finding: severity, the exact payload that triggered it, business impact, and a named ID like GALEOPS-2026-0047-C01 |
| Prioritised remediation roadmap | Fix order by engineering effort, not just by severity — what takes a day vs. what takes weeks |
| Compliance gap analysis | Where the findings and controls sit against EU AI Act and NIST AI RMF, if you need the paper |
| Verification plan | How to prove each finding is closed, and when to re-test |
That list is the product. The price buys the hours it takes to do all seven items on your agent, not a PDF template.
The sample report I publish walked one production-style LLM customer-support agent end to end. It came back with 1 Critical, 3 High, 4 Medium, and 2 Low findings. Three of them are worth naming because they're the ones I find on most agents:
The report also flagged quieter things: a broken output filter (homoglyph substitution walked right through it), no rate limiting on the chat endpoint, and file-upload OCR processing with no sandbox. None of these are model problems. All of them are the same software-security problems I've been fixing for 14+ years in enterprise IT, wearing a new protocol.
The market at the bottom end is $500–$3,000, and at that price you're usually getting one of two things: an automated scanner report, or a checklist review. Both share the same tells —
None of this means expensive audits are automatically better. It means the deliverable list is the price. A $50K engagement that skips tool misuse is worth less than a $6K one that includes it.
The $6,000 audit is the full engagement. If you want a taste of the methodology first, the $149 Snapshot runs 25+ targeted probes against one public endpoint and delivers a branded PDF plus a walkthrough video in 48 hours. And the $750 Risk Review deepens it — its full fee is credited toward the $6,000 audit if we move forward. You don't have to bet $6K on the first interaction.
A mapped attack surface, manual injection and tool-misuse testing, a findings report with reproductions and named finding IDs, a prioritised remediation roadmap with effort estimates, and (where wanted) an EU AI Act / NIST AI RMF gap analysis. Everything above is in the published sample report.
Five business days from agreed scope and access. The remediation work is yours; for the sample report's Critical and High findings I estimated 3-5 engineering days, with the guardrail-layer changes needing 2-3 weeks.
No. Scanners are good for breadth on the prompt layer, and I use tooling where it helps. But tool permission abuse, chained tool calls, and vector-DB leakage need a human who understands your business logic.
You still get the map, the tested controls that held, and the risks to monitor as your agent changes. A clean result is evidence for your board, not a wasted engagement.
We'll map your agent's attack surface together and see where it stands. No cost, no commitment.
Book the fit call →
Read next: AI Security Audit Pricing: Why a $6K Audit Is Worth More Than a $50K One · What Is an AI Security Audit? A Practical Guide
Or try the free AI Prompt Scanner · Read the sample report · ← Back to all posts