AI Governance FAQ: Straight Answers for Enterprise Teams
Short, sourced answers to the questions teams ask most when they set up AI governance, with links to the full guides.
Answer first
Most AI governance questions come down to four things: what you run, who owns it, what evidence each system needs before it ships, and who can stop it. The answers below are deliberately short. The guides below go deeper, and the primary sources behind the answers are listed at the end of the page.
Guides behind these answers
- Free tools: AI risk-tier classifier and AI acceptable use policy starter
- Setting up governance: Enterprise AI governance that unblocks delivery
- Agents and permissions: Access control patterns for enterprise agents
- Making rules enforceable: Policy-as-code for AI systems
- Staff rules: What to put in an AI acceptable use policy
- Banking and model risk: Model risk management for business AI after SR 26-2
- Standards: ISO/IEC 42001 vs NIST AI RMF and AI governance frameworks and standards, verified
- Dates: AI compliance dates to track
- Leadership: What boards need to hear about AI programs
How we keep these answers current
Regulatory dates and standards change. Each answer was last checked against its primary source on 7 November 2026, and the page is updated when a source changes.
FAQ
- What is AI governance?
- AI governance is the set of decisions, owners and controls that determine which AI systems an organization runs, what data and permissions they have, what evidence is required before release, and who can stop them. In practice it is an inventory, risk tiers, named owners and checks in the delivery process.
- What should an AI governance framework include?
- At minimum: an inventory of AI systems, risk tiers based on what each system can read and do, a named business and technical owner per system, evidence requirements per tier (evaluations, permission reviews, logging, rollback), a review cadence, and a reporting line to leadership.
- Who owns AI governance in a company?
- Leadership owns the overall system and its risk appetite. Each AI system then needs a named business owner, accountable for outcomes, and a technical owner, accountable for behaviour. A small cross-functional group usually approves high-risk releases. 'The AI team' on its own is not an owner.
- What is an AI inventory and why do we need one?
- An AI inventory is a single register of AI systems in use or in development, with purpose, owner, model or vendor, data access, actions, users affected, risk tier and last review date. Without it you cannot tell what you are governing. US banking guidance (SR 26-2) still describes a comprehensive model inventory as common practice.
- How do you classify AI risk?
- Classify by consequence and capability rather than by model. A simple scale: assistive (drafts for a human), decision support (feeds a decision a human approves), acting (writes or sends without per-item approval), and regulated or high impact (uses covered by law, such as EU AI Act high-risk categories). Tool access and write permissions raise the tier fastest.
- How often should AI systems be reassessed?
- Set a cadence by tier, for example yearly for assistive tools and quarterly for systems that act, and require reassessment on any material change: a new model version, new data source, new tool or new user group.
- What is shadow AI?
- Shadow AI is AI use outside the approved tools and processes, such as personal accounts on chat assistants or AI features switched on inside software without review. An effective response is a published list of approved tools and a fast request path, backed by discovery through SSO, expense and network data.
- Is the NIST AI RMF mandatory?
- No. NIST describes AI RMF 1.0, released in January 2023, as intended for voluntary use. Customers, contracts or regulators may still ask you to show alignment with it. NIST has said the framework is being revised.
- Can an organization be certified against ISO/IEC 42001?
- Yes. ISO/IEC 42001:2023 specifies requirements for an AI management system, and accredited certification bodies audit and certify organizations against it.
- Does ISO/IEC 42001 certification mean EU AI Act compliance?
- No. A certified management system helps organize the work, but the EU AI Act sets specific legal obligations based on your role and the system's risk category, and those must be met directly.
- When do EU AI Act high-risk rules apply?
- According to the European Commission, after the 2026 AI Omnibus amendments the rules for Annex III high-risk uses apply from 2 December 2027, and for high-risk AI built into products covered by Annex I from 2 August 2028.
- Does SR 26-2 cover generative AI or AI agents?
- No. The SR 26-2 attachment says generative AI and agentic AI models are not within its scope; its principles apply to traditional quantitative models and non-generative, non-agentic AI. Organizations are expected to govern generative and agentic systems through their own risk management practices.
- Who should approve a new AI use case?
- The system's business owner approves the use case. Low-risk assistive uses can follow a standard path with no committee. Systems that act on records, customers or money, or that fall under regulation, should go to a small group with authority to approve, with a published turnaround target.
- What are the main security risks to govern in LLM and agent systems?
- The OWASP Top 10 for LLM Applications (2025) is a good checklist. Prompt injection and excessive agency matter most for agents, because they turn hidden instructions in content into real actions.