ACF v1.0 — Now Open for Registration
Behavioral Trust is the New Standard.
The Agent Certification Framework independently verifies AI agent behavior across 30 rigorous tests — commitment boundaries, consistency, hallucination resistance, and adversarial robustness.
30
Behavioral Tests
4
Test Suites
5
Industry Modules
4
Certification Tiers
ACE Trust Network
Independent certification infrastructure for deployed AI agents
Independent
No conflicts of interest
Public Registry
Verifiable in real time
Standards-Based
Open framework, public comment
EU AI Act Aligned
ACF-EU module available
Working Group Governed
Community-governed standards
You can’t trust what you can’t verify.
30
Behavioral tests per cycle
4
Suites covering every failure mode
5
Industry-specific overlay modules
100%
Public registry — every cert verifiable
AI agents are being deployed into consequential roles — customer interactions, financial decisions, healthcare triage, legal research. Yet most operators have no independent evidence that their agent behaves as designed under adversarial conditions, edge cases, or extended sessions.
ACF exists to change that. We provide a rigorous, independent, public certification that any stakeholder can verify in seconds — buyers, regulators, boards, and end users.
Behavioral, Not Just Technical
We test what the agent actually does — not just what the underlying model can do. Configuration, tooling, and prompting are all in scope.
Adversarial by Design
Every certification includes deliberate manipulation attempts. If your agent breaks under adversarial pressure, we find it before your users do.
Publicly Accountable
Certificates are published to the public registry with scores, tier, and issue date. No black-box badges — every claim is verifiable.
Four suites. Every failure mode.
ACF v1.0 organizes 30 behavioral tests into four named suites. Each suite targets a distinct class of agent failure. Higher tiers include all lower-tier suites cumulatively.
Full Standards Documentation →Suite CB · 8 Tests
Commitment Boundary
Tests whether the agent respects its defined operational scope and refuses to act outside authorized boundaries — including privilege escalation, tool misuse, and confidentiality violations.
Suite CS · 7 Tests
Consistency
Tests whether the agent produces stable, predictable behavior across equivalent inputs, extended sessions, and varying user framings — including policy uniformity and internal contradiction resistance.
Suite HD · 8 Tests
Hallucination Detection
Tests whether the agent produces fabricated, ungrounded, or misleading outputs — spanning citation accuracy, numerical claims, entity confabulation, and appropriate uncertainty expression.
Suite AR · 7 Tests
Adversarial Resistance
Tests whether the agent maintains behavioral integrity under deliberate manipulation — including prompt injection, jailbreak attempts, social engineering, context poisoning, and indirect injection via retrieved documents.
From submission to badge
Four stages, fully transparent. Operators receive detailed score reports at every stage.
Submission
Submit agent documentation, API endpoints, and system prompt scope via the operator portal. Schema validation and data sanitization performed at intake.
Automated Testing
Proprietary distributed testing infrastructure executes all 30 behavioral tests across designated suites — including logic fuzzing, edge-case probing, and safety benchmarks.
Expert Review
Accredited ACF auditors conduct red-team analysis and ethical boundary evaluation on flagged outputs. Human judgment on every borderline result.
Certification
A cryptographically signed certificate is issued and published to the public registry with scores, tier, operator name, and real-time verification link.
Built for regulated industries.
Available with Tier 3 and Tier 4 certifications. Each module adds industry-specific behavioral tests and produces a regulatory evidence package.
ACF-FIN
Financial Services
Fiduciary advice boundaries, regulatory compliance, and consumer protection behavior. Aligned with SEC Reg BI, FINRA Rule 2111, MiFID II.
ACF-HLTH
Healthcare
HIPAA-adjacent behavioral controls, medical advice boundary enforcement, PHI handling. Aligned with HIPAA Privacy Rule, AMA AI ethics guidelines.
ACF-LEGAL
Legal Services
Unauthorized practice of law detection, privilege boundary handling, conflict-of-interest avoidance. Aligned with ABA Model Rules, state UPL statutes.
ACF-GOV
Government & Public Sector
Accessibility, auditability, and non-discriminatory output requirements. Aligned with Section 508, NIST AI RMF, OMB AI guidance.
ACF-EU
EU AI Act Alignment
Tests mapped to Articles 9, 10, 11, 13, 14, and 15. Produces a conformity evidence package supporting CE marking efforts. Note: ACF-EU is an alignment-supporting module — not a substitute for notified body conformity assessment under Annex VII of the EU AI Act.
Built for enterprise procurement.
Procurement and vendor management teams use ACF certification as a prerequisite before AI vendor conversations begin. ACF gives procurement the independent evidence it needs to move deals from legal review to signature.
Pre-qualification
Require Tier 2+ certification before a vendor enters your procurement process
Due Diligence
Review full score reports as part of technical evaluation
Ongoing Monitoring
Confirm annual recertification status as part of vendor review cycles
Third-Party Viewer Access
Grant named procurement contacts view-only access to full score reports without making them fully public ($150/yr add-on)
The ACF verification API allows programmatic certificate status checks. Integrate directly into vendor management platforms, procurement portals, and compliance dashboards.
Verify a Certificate →What operators say
“ACF’s behavioral testing surfaced failure modes our internal QA team had completely missed. It’s not just a badge — it’s a structured audit that made our agent meaningfully safer.”
“The adversarial resistance suite identified a prompt injection vector in our retrieval pipeline. Our security team had run a red-team exercise two weeks prior and missed it entirely.”
“Enterprise buyers now ask for ACF certification before procurement conversations. It’s become the proof point that moves deals out of legal review and into signatures.”
Design Partner Program
ACF’s founding cohort of early operators. 10 slots. Once filled, the program closes permanently.
7 slots remaining.
Benefits
- Free Tier 1 + Tier 2 certification ($1,098 value)
- 30% permanent discount on all future certifications
- Founding listing in the public registry with “Design Partner” badge
- Early access to new test suites and module previews
- Input rights on standard revision proposals
- Priority auditor assignment — no queue wait
Obligations
- Complete one full certification cycle within 90 days of acceptance
- Provide structured feedback (survey + optional 30-min call)
- Allow ACF to publicly reference them as a founding design partner (with approval rights)
EU AI Act Alignment
The ACF-EU module is available with Tier 3 and Tier 4 certifications. It produces a structured evidence package mapped to EU AI Act requirements for high-risk AI systems.
Article 9
Risk Management System
Article 10
Data and Data Governance
Article 11
Technical Documentation
Article 13
Transparency and Provision of Information
Article 14
Human Oversight
Article 15
Accuracy, Robustness and Cybersecurity
Important: ACF-EU is an alignment-supporting module designed to produce evidence that supports conformity assessment efforts. It is not a substitute for notified body conformity assessment for systems subject to mandatory third-party assessment under Annex VII of the EU AI Act.
Common questions
What does ACF actually test?
ACF tests the deployed agent — including its system prompt, tool configuration, retrieval pipeline, and operating context. We run 30 behavioral tests across four suites: Commitment Boundary (scope violations), Consistency (behavioral stability), Hallucination Detection (fabricated outputs), and Adversarial Resistance (manipulation attempts).
How is this different from model benchmarking?
Model benchmarks test what a model is capable of. ACF tests what your deployed agent actually does. Two agents built on the same model can have completely different behavioral profiles depending on system prompt, tools, and configuration. ACF certifies the agent as deployed — not the model in isolation.
What do I need to submit?
The standard intake package requires: agent documentation (purpose, scope, deployment context), API endpoint access for live testing, system prompt scope declaration, and operator identity verification. Tier 3/4 registrations additionally require a signed Operator Registration Agreement.
How long does certification take?
Tier 1: 14 business days. Tier 2: 10 business days. Tier 3: 12 business days. Tier 4: 8 business days. Expedited processing (40% surcharge) reduces turnaround to 3–5 business days for all tiers.
Does ACF certification satisfy EU AI Act requirements?
The ACF-EU module produces evidence packages that support conformity assessment efforts under the EU AI Act. ACF is not a notified body. For high-risk AI systems requiring mandatory third-party conformity assessment, ACF documentation serves as supporting evidence but does not substitute for a notified body assessment.
What happens if my agent fails a test?
You receive a detailed score report identifying the specific failure type and test ID. You may resubmit after remediation. The first resubmission within 90 days is included in the original certification fee. Subsequent resubmissions are charged at 30% of the tier price.
Set the standard. Prove your agent.
Join the network of certified agents. Initiate your behavioral audit and secure your agent’s public registry status.