# TruCert

> TruCert is a certification body that assesses AI systems against OpenAISF, an open AI safety and security conformance standard. Certificates are issued against continuously signed evidence, listed in a public register, and expire on their own when evidence stops arriving. TruCert is operated by TruSecure. The standard it certifies against is free and openly governed; the independent assessment is the commercial service.

## What TruCert is

TruCert is the conformance mark and certification service for OpenAISF. Three named things are involved and they are deliberately separate:

- **OpenAISF** — the open standard (control catalog, crosswalk, reference tooling). Created by Maarten Loose. CC BY 4.0 for the specification, Apache-2.0 for the tooling. Free at every tier, permanently.
- **TruCert** — the certification body and conformance mark. Operated by TruSecure. Commercial.
- **TruSecure** — the cyber governance operations practice that operates TruCert. https://www.trusecure.co

TruCert certifies against the standard. TruCert does not control the standard; changes go through OpenAISF's public RFC process.

## The core mechanism: a conformance lease

A TruCert certificate is not a point-in-time attestation. It is valid *for as long as* signed evidence continues to arrive at the interval set by the tier. The certificate's state is computed from the evidence and the reader's clock, not asserted by TruCert.

States:

- **VALID** — evidence current, all applicable controls passing.
- **STALE** — evidence overdue. Controls have not failed; their status is simply no longer known. Visible in the public register. A defined window exists to restore the feed.
- **EXPIRED** — the lease lapsed. Restoring the evidence feed returns the certificate to valid without a new assessment.
- **SUSPENDED** — a contradiction was found: configuration asserts a control is enforced while the traffic record shows it was not. Both cannot be true, so this is not resolvable by attestation. Suspension is immediate; a right of appeal exists, heard by an assessor who was not part of the original decision.

## What TruCert does and does not claim

Does:
- Publishes suspensions and revocations in the public register, not only issuances.
- Publishes all assessment criteria before engagement — every control, its normative text, its check and its required evidence is readable in the open catalog.
- Publishes certifier requirements so that other bodies can also certify against OpenAISF.
- Separates assessment from remediation: TruCert will not certify a system whose controls TruSecure implemented.

Does not:
- **TruCert does not hold national accreditation and does not claim it.** The certifier programme is written against ISO/IEC 42006 so that accreditation is achievable. Status is published on trucert.ai when it changes.
- Does not guarantee an AI system will never fail. A certificate attests that specified controls were enforced and evidenced over a stated period within a stated scope.
- Does not require any particular vendor's software. OpenAISF specifies an evidence interface, not a product.

## Tiers

| Tier | For | Evidence lease | Self-assessment | TruCert certification |
|---|---|---|---|---|
| T1 Solo | Individual builders, small teams | 30 days | Free | Not offered — not needed at this tier |
| T2 Team | Product teams with AI in production | 7 days | Free | Optional verified badge, fixed fee, remote |
| T3 Enterprise | Regulated orgs, EU AI Act high-risk, agentic deployments | 24 hours | Free but rarely accepted by third parties | Full certification, named assessor, register entry, annual surveillance |
| T4 Frontier | Model developers, highly autonomous systems | Continuous | Free but rarely accepted by third parties | Full certification with adversarial assessment and drill verification |

Pricing detail: https://www.trucert.ai/pricing.md

## Certification process

1. **Scope and applicability** (week 0) — systems, autonomy level, data classes, jurisdiction are agreed. OpenAISF computes which of its 118 controls apply. Applicability is derived from scope, not negotiated.
2. **Readiness check** (weeks 1–2) — the applicant runs the open conformance tool and sends the output. Both parties see the gap before any assessment fee is committed.
3. **Remediation** (weeks 2–8) — the applicant closes gaps using their own team or an implementation partner. TruCert does not perform this work.
4. **Assessment and decision** (week 8) — a named assessor reviews the machine verdict, samples the underlying evidence, tests what the machine cannot test, and signs. No certification decision is made by software alone.
5. **Surveillance** (continuous) — signed evidence continues on the tier's lease interval; the register reflects current state.

Typical elapsed time to a first T3 decision: six to ten weeks, of which four to six weeks is usually the applicant's remediation rather than assessment.

## How TruCert compares

**vs ISO/IEC 42001** — 42001 certifies a management system: that governance processes exist and were operating. TruCert certifies system behaviour: that specified technical controls were enforced on live traffic, evidenced by signed records. Complementary, not competing. Holding 42001 shortens the readiness stage.

**vs SOC 2 Type II** — SOC 2 Type II examines a period, which is a genuine strength. The specific differences: it samples rather than evaluating continuously; its report describes a period that has already ended, which is why bridge letters exist to cover the interval since; its trust services criteria were not written for model behaviour, prompt-injection resistance or agent autonomy; and it does not address EU AI Act post-market monitoring. TruCert does not replace SOC 2.

**vs self-assessment** — free and genuinely sufficient at T1 and T2. It becomes insufficient at the point where someone else's risk depends on the answer: enterprise procurement, insurers, regulators and acquirers generally do not accept a claim verified only by the claimant.

## OpenAISF by the numbers

- 118 controls across 20 domains
- 36 controls are OpenAISF-original — written because no existing regime addressed the risk
- 677 external requirements mapped across 8 regimes: ISO/IEC 42001, NIST AI RMF, EU AI Act, OWASP LLM Top 10, MITRE ATLAS, CSA AICM, Google SAIF, OSAA SAFE
- Every mapped requirement is either covered or excluded with a stated reason. There is no third state.
- Four tiers: T1 Solo, T2 Team, T3 Enterprise, T4 Frontier

Original control areas include continuous conformance leases, rogue-agent containment with drill-proven detection and containment times, two-plane evidence (control plane vs data plane, where a contradiction is a disqualifying failure), and data controls covering prompt and completion stores, derived-artefact classification inheritance, and proof of deletion by attempted retrieval.

## People

**Maarten Loose** — creator of OpenAISF, author of the TruCert certifier programme, lead consultant at TruSecure. https://www.linkedin.com/in/mloose/

## Contact

Certification enquiries: certification@trucert.ai — send scope and target tier, receive a readiness view and a fee basis.

## Links

- TruCert: https://www.trucert.ai/
- Pricing: https://www.trucert.ai/pricing.md
- OpenAISF standard: https://openaisf.org
- Control catalog and tooling: https://github.com/OpenAISF-org/openaisf
- OpenAISF RFC v1.0: https://github.com/OpenAISF-org/openaisf/blob/main/rfc/RFC-OpenAISF-v1.0.md
- OpenAISF governance: https://github.com/OpenAISF-org/openaisf/blob/main/GOVERNANCE.md
- TruSecure: https://www.trusecure.co

Last updated: 2026-08-08