AIUC-1 vs Other AI Governance Certifications

AI agents don't just generate content anymore. They execute transactions, call external tools, handle sensitive customer data, and make sequential decisions that chain together without a human checking each step. That shift, from AI-as-tool to AI-as-actor, changes what can go wrong and how badly. A chatbot that hallucinates a wrong answer is embarrassing. Prompt injection, data leakage, and the brand exposure that comes from an agent saying something it shouldn't in front of a customer push the failure modes toward looking less like software bugs and more like operational risk.

Existing governance frameworks were designed to evaluate organizations and management systems. None of them were built to test whether a specific agent holds up under adversarial pressure in production.

That gap is not just theoretical anymore. The EU AI Act's phased rollout has made it a compliance deadline Workstreet Medium. Prohibited practices and AI literacy obligations became enforceable in February 2025, governance and general-purpose AI obligations followed in August 2025, and Commission enforcement powers begin in August 2026, with general application arriving the same month and high-risk system deadlines pushed to December 2027 and August 2028 under the Digital Omnibus Workstreet Medium. The global AI regulatory compliance market is projected to hit $15 billion by 2026, which is a fairly blunt signal that governance has stopped being optional and started being a line item Workstreet Medium.

So what's actually missing? Organizations don't just need to prove their AI management system is well-designed on paper. They need a way to show that a particular agent, the one actually touching customer funds or patient records, behaves safely once it's live. That's a narrower and harder question than "does this company have good AI policies," and it's the question none of the existing frameworks were built to answer directly.

AIUC-1, how it works, and where its authority comes from

AIUC-1 was launched in 2025 by the Artificial Intelligence Underwriting Company, a venture-backed startup, with development input from people who worked at Anthropic and institutional partners including Orrick, Stanford, the Cloud Security Alliance, MIT, and MITRE. The company describes it as "SOC 2 for AI agents," a comparison that says a lot about the ambition behind it: something vendible, auditable, and verified by a third party, specifically at the level of an individual agent rather than an entire organization.

The standard itself runs more than 50 controls spread across six domains: Safety, Security, Reliability, Accountability, Data & Privacy, and Society, mapped against MITRE ATLAS and the OWASP Top 10 for Agentic Applications What to Make of AIUC-1, a New AI Agent Certification. Certification requires an annual audit from an accredited auditor. Schellman was the first firm accredited to conduct these audits, and ElevenLabs became the first voice AI company certified under the standard. The standard itself moves fast too. The January 2026 update alone touched 26 existing requirements and added more than 40 voice-specific ones. Vendors seeking certification go through more than 2,000 technical evaluations covering enterprise risk scenarios, and the framework's development involved input from a consortium of more than 250 Fortune 500 security leaders UiPath What to Make of AIUC-1, a New AI Agent Certification.

Now, the caveat. ISO 42001 operates through internationally accredited certification bodies. SOC 2 sits under the AICPA's authority. NIST frameworks carry the weight of a federal agency. AIUC-1 is different: AIUC itself accredits the auditors who certify against its own standard. The framework's authority rests on AIUC's own standing rather than an outside accreditation body, a point raised directly in Lenny Zeltser's analysis of the standard. Does that matter? Zeltser raises an incentive-chain question that's hard to wave away. AIUC writes the framework, runs the technical evaluations, issues the certificates, and also sells AI agent insurance, a vertical integration that echoes the issuer-pays dynamic that's long drawn criticism in credit ratings. AIUC's counter is that its own insurance exposure on certified agents creates a direct financial incentive to keep the bar high; if a certified agent causes a loss, AIUC is the one paying out. Whether that counter-incentive is strong enough to offset the conflict is something the market will judge over time, not something settled by the argument alone.

One more wrinkle: AIUC-1 doesn't define what counts as an "AI agent." The vendor decides. The vendor also decides which specific agent gets submitted for certification. Buyers evaluating a certified vendor should ask exactly which agent, which tools, which model versions, and which data flows the audit actually covered, because "AIUC-1 certified" can describe a narrow slice of a much larger product. None of this disqualifies the standard. It just means the standard is young enough that its governing structure and scope conventions are still being tested in public, the way any new certification is in its first few years. Certificates remain valid for 12 months, but quarterly technical retests are required to maintain status Workstreet.

ISO/IEC 42001 and AI governance at the organizational level

ISO/IEC 42001:2023 is the first global standard for an AI Management System, covering how an organization establishes, runs, and keeps improving its governance of AI across its operations. Expect ISO 42001 to follow that same trajectory as more buyers add it to vendor questionnaires.

ISO 42001 addresses how an organization governs AI (policies, processes, oversight structures, and risk management across the AI lifecycle). AIUC-1 asks something narrower and more technical: does this specific agent's safeguards hold up when someone tries to break them. One is organizational. The other is use-case level.

Schellman, which audits against both standards, offers sequencing guidance that's genuinely practical. An organization with mature AI governance but nothing formalized should pursue ISO 42001 first, then layer in AIUC-1's technical controls. An organization that already has strong technical safeguards but lacks the surrounding governance structure should do the reverse, AIUC-1 first, ISO 42001 second, to build the scaffolding around what already works. UiPath holds both certifications. So does Fieldguide, which also carries a SOC 2 Type 2 report, and that combination is quickly becoming something close to the expected baseline for enterprise AI vendors. The practical read is straightforward: ISO 42001 builds the governance framework, AIUC-1 checks whether the technical safeguards inside that framework actually function under pressure. Neither one makes the other redundant. It is certifiable through internationally accredited certification bodies, with a certificate valid for 3 years and recognized globally in procurement, and the ISO 27001 precedent suggests ISO 42001 is becoming a procurement gate in the same way. The distinction from AIUC-1 is structural.

What NIST AI RMF offers

It's widely used internally as a structuring tool for risk programs. What it doesn't offer is a certification path. NIST is explicit that the framework is meant to be tailored to an organization's context, not audited against by a third party for a certificate.

NIST has extended into more specific territory with its Cyber AI Profile, IR 8596, which addresses where cybersecurity and AI risk intersect. A draft was released in 2025, and while the subject matter runs adjacent to what AIUC-1 covers, it's guidance rather than a competing certification. Interestingly, AIUC-1 explicitly maps its own controls back to the NIST AI RMF. A vendor holding AIUC-1 certification can point to that mapping as evidence that NIST's guidance has been not just adopted but independently tested. Where NIST fits is as an indispensable tool for internal risk program design, but a limited trust signal to external stakeholders.

That mapping matters because of what NIST alone can't provide. NIST guidance produces internal documentation. It doesn't produce third-party validation. NIST AI RMF 1.0 is voluntary, sector-agnostic, and flexible (designed to help teams identify, measure, and mitigate AI risks throughout the AI lifecycle), and is widely used as an internal risk-structuring tool. The practical gap NIST doesn't fill is that buyers in procurement and RFPs need a verifiable artifact (a report, a certificate) that an auditor has tested the agent, while NIST guidance produces internal documentation, not third-party validation.

What SOC 2 covers in AI contexts

SOC 2 sits under the AICPA's Trust Services Criteria and evaluates a service organization's general controls: security, availability, processing integrity, confidentiality, and privacy. It's a well-established, well-understood standard, and for good reason. But it wasn't built with agentic AI in mind, and the gaps show. Prompt injection, unauthorized tool calls, hallucinations that turn into operational decisions rather than just bad text, adversarial manipulation of an agent's behavior, none of these were on anyone's radar when the Trust Services Criteria were written, so none of them are meaningfully covered.

Zeltser's analysis frames SOC 2 and AIUC-1 as complementary rather than competing: SOC 2 attests to the general control environment of the service organization, AIUC-1 attests to the behavior of a specific agent operating inside that environment. Fieldguide holds both, SOC 2 Type 2 and AIUC-1, and that pairing is a good illustration of how the market is actually layering assurance rather than swapping one framework for another.

For assurance professionals who already know SOC 2 cold, a lot of that experience transfers directly. Scope definition matters just as much in AIUC-1. Auditor selection matters just as much. Evidence quality matters just as much. And the same risk that threatened SOC 2's credibility in its early years, a race toward "fast and easy" auditors competing on speed rather than rigor, is a live risk for AIUC-1 too. For accounting and assurance firms sizing up AI vendors for their own practices, the practical conclusion is blunt: SOC 2 Type 2 is necessary, but it's no longer sufficient on its own when the vendor's product involves autonomous agents.

ISACA's AAIR, AAISM, and AAIA credentials, their approach to AI risk, and their intended audience

AAIR, Advanced in AI Risk, launched in April 2026 and has been picked up quickly by IT risk teams because it builds directly on CRISC and ISACA's other established certifications, focusing on AI risk management, governance, and lifecycle risk. AAISM, Advanced in AI Security Management, is about protecting AI systems from attacks, adversarial inputs, and operational threats, making it the ISACA credential closest to the actual threat model AIUC-1 tests against at the system level. AAIA, Advanced in Artificial Intelligence Audit, launched in mid-2025 and is aimed at auditors evaluating AI systems directly, which makes it particularly relevant for CPA and assurance professionals whose clients are deploying agentic AI. Training Camp's analysis lists the ISACA stack alongside ISO 42001 as carrying the strongest enterprise recognition heading into 2026.

The distinction to hold onto here is who gets certified. AAIR, AAISM, and AAIA certify a person's competence. AIUC-1 and ISO 42001 certify a vendor's system or a specific agent.

What the IAPP AI Governance Professional credential covers

The International Association of Privacy Professionals launched its AI Governance Professional credential, AIGP, in 2024, building it on top of the already well-established CIPP and CIPM ecosystem.

AIGP has become the most recognized AI governance credential inside the privacy and compliance world, though its scope is narrower than the ISACA stack. It's oriented toward policy, privacy law, and governance design, not technical risk assessment or audit methodology. That makes it the right fit for legal, compliance, and privacy teams building or advising on AI governance programs, but it's not a substitute for AAIA or AIUC-1 if the job actually calls for auditing a system or certifying a vendor's technical safeguards. AIGP is a strong entry point for a privacy professional moving into AI governance work. It just doesn't reach down into agentic risk at the control or system level, and it was never designed to. The exam format consists of 90 questions over 4 hours, costs $550, has no formal prerequisites though a background in privacy or governance is helpful, and preparation typically requires 40 to 60 hours What to Make of AIUC-1, a New AI Agent Certification.

The early adopters who have sought AIUC-1 certification

Two certifications, so far, tell most of the story. UiPath, listed on the NYSE as PATH, announced AIUC-1 certification on March 9, 2026, becoming the first enterprise automation platform to meet the standard. The certification covered Intelligent Extraction Processing, Agents, and Autopilot, tested across more than 2,000 enterprise risk scenarios UiPath. UiPath also holds ISO/IEC 42001:2023 and was a Founding Technical Contributor to AIUC-1 itself, which suggests the company was involved in shaping the standard well before it sought certification against it.

Fieldguide, an AI platform built for audit and advisory work, announced its own AIUC-1 certification on May 6, 2026, the first AI platform in the audit and advisory category to do so. Its Field Agents were the certified product, and the certification was built on top of an existing ISO 42001 certification and SOC 2 Type 2 report, an example of the layering pattern that repeatedly occurs across this space. Notably, Fieldguide's own platform supports AIUC-1 audits, meaning CPA firms using it can turn around and use it to certify their own clients' AI agents. KPMG shows up in a different way: Aisha Tahirkheli, the firm's Principal and Trusted AI Leader, is cited as an endorser of AIUC-1 on AIUC's own website, which signals enterprise professional services awareness even without a reported certification. And Schellman threads through both major certifications so far, having audited UiPath and Fieldguide as the first accredited AIUC-1 auditor while also holding standing as the world's first ANAB-accredited ISO 42001 certification body, a dual role that matters for any firm thinking about auditor selection.

What does this pattern actually say? Early adopters follow a distinct pattern. They're enterprise automation platforms and AI-native professional services tools, exactly the vendors whose products sit inside the workflows where agentic failure carries the steepest cost: financial transactions, patient data, audit engagements. For CPA and assurance firms specifically, Fieldguide's certification is the most directly relevant precedent on record. It shows that AI agents supporting actual audit work can be independently validated, and it gives firms evaluating AI vendors a concrete standard to point to in procurement conversations rather than relying on vendor claims alone.

Using these standards as decision tools matched to the organization's actual situation

None of these frameworks rank neatly above one another. They serve different purposes for different audiences, and the right one depends entirely on the problem in front of the organization asking the question.

A vendor deploying AI agents into enterprise workflows, financial, healthcare, legal, audit, gets the most direct signal from AIUC-1, with ISO 42001 covering the surrounding management system and SOC 2 Type 2 still functioning as the baseline expectation for any service organization. For an auditor or assurance professional whose clients deploy agentic AI, ISACA AAIA addresses audit methodology, while AIUC-1 is what they'll encounter in vendor assessments and potentially in client engagements.

A procurement team evaluating AI vendors sits at the intersection of all of this. The honest approach isn't picking a single favorite framework and demanding it from every vendor. It's understanding what each certification actually tests, checking scope carefully (especially with AIUC-1, given that vendors define their own agent boundaries), and treating the presence of multiple, layered certifications, ISO 42001 plus AIUC-1 plus SOC 2, the way UiPath and Fieldguide both do, as a stronger signal than any single one standing alone. For an organization building an internal AI risk program, NIST AI RMF provides the structural foundation, while ISACA AAIR or AAISM equips the practitioners who run it. For a privacy or legal team advising on AI governance policy, IAPP AIGP is the most relevant professional credential.