Old Risk Frameworks Can't Handle AI. Here Are the New Ones That Try.

Key points

• ISO/IEC 42001, published December 2023, is the first certifiable international management system standard for AI.
• NIST AI RMF, released January 2023, is free and public, built around four functions: Govern, Map, Measure and Manage.
• ENISA's FAICP, published June 2023, runs across three progressive layers and anticipates the EU AI Act.
• The frameworks are complementary: different intents, different gaps, often deployed together.
• ISO 42001's full text is paywalled, which is a concrete barrier for smaller organizations early in their AI governance journey.

Decades of enterprise risk management doctrine didn't anticipate adversarial inputs or AI supply chain poisoning. The frameworks that followed are still young.

A new generation of AI-specific standards has arrived, and none of them solve the same problem. Some tackle governance and organizational accountability; others address technical security controls or regulatory alignment. Picking the right combination depends on where your gaps actually are, as our 28 May piece on AI governance living outside the pipeline made plain.

What does ISO/IEC 42001 actually require?

Published in December 2023, ISO/IEC 42001 is the first internationally recognized formal management system standard for AI, mirroring the structure of ISO 27001. Organizations must document how they design, monitor and control AI systems, conduct AI impact assessments, and demonstrate governance over third-party suppliers and data pipelines.

It is voluntary and certifiable, applying across sectors. A growing number of organizations use it to show alignment with the EU AI Act. Nicole Carignan, CISO at Darktrace, calls it the strongest foundation for building an AI risk program. "It forces organizations to think holistically about ownership, governance, oversight, data integrity, security risk mitigation, accountability and continuous improvement," she says. The downside: it is resource-intensive, and the full standard is paywalled, a real barrier for organizations early in their AI governance journey.

Should you start with NIST instead?

Released in January 2023, the NIST AI Risk Management Framework is public and free, built around four interconnected functions: Govern, Map, Measure and Manage. It does not hand out pass/fail grades. That matters.

Ram Varadarajan, CEO at Acalvio, recommends it as a first stop because "it forces the three conversations that have to happen first: who owns AI risk, what AI is actually running, and who gets hurt if something goes wrong." Forrester analysts welcomed the framework but flagged conflicts of interest among its drafters, a missing data governance layer, and its descriptive rather than prescriptive character. Chief data officers applying it need to interpret carefully.

Where does ENISA's FAICP fit?

The European Union Agency for Cybersecurity published its Framework for AI Cybersecurity Practices in June 2023, designed to anticipate the EU AI Act. It runs across three layers: foundational IT security practices, AI-specific risks including adversarial attacks and model tampering, and sector-specific guidance for energy, healthcare and telecoms.

FAICP is voluntary, but EU regulators treat it as a governance baseline for organizations operating inside the bloc. Varadarajan expects it to follow the path of Europe's data privacy law, becoming the reference point for companies worldwide regardless of where they're headquartered.

Carignan's read is the most useful one for practitioners: the overlap across these frameworks reinforces the core practices that matter. Governance and accountability don't change. The attack surface does, fast.