Baseline
$6,500
5 business days
One agent workflow, all 51 scenarios, one model, and a score compared across judges.
AgentBoundary Certified
Independent robustness testing for AI agents.
Adversarial testing against 51 prompt-injection scenarios, execution-based scoring, and a verification package using the open AgentBoundary benchmark.
Certification paths
Go directly to the certification task you need.
Who it is for
AI SaaS vendors, security teams deploying internal copilots and agents, and LLM/MLOps platforms that need an independent report on a defined agent workflow.
When to certify
Use this before enterprise security review, before granting production access to an agent, or when a model upgrade has invalidated prior testing.
Why certify
Give reviewers test results they can assess.
- An agent may reach production before its permissions and instruction boundaries have been tested under adversarial conditions.
- Security reviewers need test methods, results, and limitations they can examine alongside a vendor’s security claims.
- Model swaps and prompt changes can alter behavior, so prior results need to be reassessed when the tested system changes.
Certification tiers
Three tiers, from a single-flow baseline to a broader review with custom adversarial scenarios and board reporting.
Production
$18,000
10 business days
Up to three agent workflows, all four scenario types (RAG, filesystem, browser, code), comparison across judges, one rerun after fixes, and a procurement memo.
Board-Ready+
$38,000
15 business days
Up to six workflows, the full suite plus custom attack scenarios, an executive briefing, a remediation workshop, and optional quarterly recertification.
Annual recertification costs 55–60% of the initial tier. Targeted retesting after a major model change is scoped to the affected workflows.
What you receive
A certification package built for security and procurement review.
Boundary Robustness Scorecard
Attack-instruction compliance rates overall and by attack family, Wilson confidence intervals for each rate, and a 0–100 AgentBoundary Robustness Score.
Technical Report
Failing prompts, execution traces, recurring causes, and remediation tasks ranked by severity.
Procurement Appendix
Test methods, comparison of judge results, assessed workflows and versions, limitations, and a signed attestation letter.
Public Verification Badge
12-month verification listing with tier, certification date, verification URL, and revocation terms. Confidential by default; the vendor controls publication.
Recertification Plan
Annual recertification and targeted retesting after material model or architecture changes. Each result applies to the recorded scope, versions, and test date.
Methodology
Documented methods and scoped results.
The review uses the AgentBoundary benchmark. Reports document the tested workflow, scenario set, scoring method, and limitations so reviewers can understand how each result was reached.
- Execution-based scoring: repeating injected text alone does not count as following its instructions.
- Cross-judge comparison across three architecturally distinct judges to reduce reliance on any one model’s judgments.
- Wilson score confidence intervals for every binomial compliance rate.
- Attack families covered: markdown injection, cognitive overload, RAG poisoning, context drift, and HTML injection.
- Agent scenario types covered: RAG, filesystem, browser, and code agents.
Hard questions, answered
“We already run the open-source benchmark, or our internal red team tests this.”
Independent testing can complement your internal review with a documented scope, recorded results, and an audit trail that customer security teams can examine.
“LLM-as-judge scoring is subjective, or could be wrong.”
We use execution-based scoring with structured rubrics, multiple judges, and Wilson score confidence intervals on measured compliance rates. Model judgments can still be wrong; results must be read with the test scope and limitations.
“A bad score could be used against us.”
All results are confidential. You have sole discretion over whether and when to display the public badge. A private pre-certification window is included; publish only after a remediation pass clears your chosen threshold.
How results are supported
The open benchmark provides the scenarios. The certification package adds documented test methods, execution-based scores, signed validation materials, and records for security and procurement review.
Certification scope
Certify your agent before your next enterprise security review.
Apply to start with a short scoping call. Turnaround, remediation windows, and any public verification options are confirmed in scope.