AI Compliance Frameworks

The AI compliance landscape shifted from voluntary best practices to mandatory regulatory requirements faster than most organizations anticipated. Understanding which frameworks apply to your AI deployments — and what each requires — is now a core governance responsibility. This page explains the four most important frameworks in operational terms: what they are, who they cover, and what you actually have to do.

NIST AI Risk Management Framework (AI RMF)

The NIST AI RMF is structured around four core functions that together form a lifecycle approach to AI risk management:

• GOVERN: Establish the organizational context for AI risk — policies, roles, responsibilities, culture. This is where you define who owns AI risk and how decisions get made.
• MAP: Identify and categorize AI risks in context. What is this AI system doing? What could go wrong? Who is affected? What are the impacts if it fails or is compromised?
• MEASURE: Analyze, assess, and track AI risks. This is where you actually test and evaluate your AI systems — including for security, fairness, and reliability.
• MANAGE: Prioritize and act on identified risks. Implement controls, monitor for new risks, and have response plans for when things go wrong.

The companion document — the AI RMF Playbook — provides specific suggested actions for each function. It's available free from NIST and is the most actionable starting point for organizations building an AI risk program aligned to this framework.

EU AI Act

The EU AI Act uses a four-tier risk classification system. Where your AI systems fall determines your compliance obligations:

Tier 1: Unacceptable Risk (Prohibited)

These AI applications are banned entirely under the EU AI Act. They include: social scoring systems that evaluate people based on behavior or personal characteristics; real-time remote biometric surveillance in public spaces by law enforcement (with narrow exceptions); AI that exploits vulnerable groups; and AI that manipulates people through subliminal techniques. If your AI does any of these things, you must discontinue it regardless of business case.

Tier 2: High Risk (Strict Requirements)

This is where most enterprise AI compliance work happens. High-risk AI includes: AI used in hiring, promotion, and employee management; AI that influences credit, insurance, or financial access decisions; AI used in critical infrastructure (energy, water, transport); AI used in education and vocational training; AI used in essential services; and AI used in law enforcement and judicial administration.

For high-risk AI, you must implement:

• A risk management system documented and maintained throughout the AI system's lifecycle
• Data governance — training data must be appropriate, relevant, and free of harmful bias
• Technical documentation sufficient for regulators to assess compliance
• Logging and traceability — the system must keep logs that allow post-deployment review
• Transparency — users must know they're interacting with an AI system
• Human oversight — the system must be designed to allow humans to understand, monitor, and intervene
• Accuracy, robustness, and cybersecurity requirements
• Conformity assessment before deployment (self-assessment or third-party, depending on category)

Tier 3: Limited Risk (Transparency Requirements)

AI systems with limited risk — primarily chatbots and systems that generate synthetic content — must disclose that they are AI systems. Chatbots must identify themselves as AI. Deepfakes must be labeled. These requirements are already in effect.

Tier 4: Minimal Risk (Few Requirements)

Most AI falls here — spam filters, recommendation engines, AI features in standard software. Few specific obligations apply, though providers may voluntarily adopt codes of conduct.

ISO/IEC 42001 (AI Management System)

ISO/IEC 42001 provides the framework for a complete AI management system. Its structure mirrors ISO 27001 (information security management) and ISO 9001 (quality management), which means organizations already certified to those standards have a significant head start. The standard covers the full AI lifecycle from planning and design through deployment, operation, and decommissioning.

The strategic value of ISO/IEC 42001 certification: it provides a credible, audited, third-party-verified signal of AI governance maturity. For enterprise B2B sales, public sector procurement, and regulated industry relationships, certification is increasingly becoming a table-stakes requirement. Organizations that pursue it proactively gain a competitive advantage. Those that wait find themselves scrambling to satisfy procurement questionnaires with evidence that doesn't exist yet.

SOC 2 + AI

SOC 2 auditors are increasingly asking questions specifically about AI systems. They want to understand: Are AI systems included in the security boundary? How is AI model access controlled? How is AI output accuracy monitored? How are AI incidents detected and responded to?

Organizations pursuing SOC 2 should proactively include their AI systems in scope rather than waiting for auditors to raise questions. The five trust service criteria apply directly:

• Security: AI systems must have access controls, vulnerability management, and monitoring — same as any other system
• Availability: AI service dependencies (API providers, model hosting) must be part of uptime and resiliency planning
• Confidentiality: Data sent to AI systems must be handled with the same confidentiality controls as other sensitive data
• Processing Integrity: AI output quality and accuracy must be monitored — you must know if the AI is producing garbage
• Privacy: Personal data processed by AI systems must satisfy the same privacy controls as other personal data processing