Loading
Skip to content
Miracle Minds AI
  • Try “HIPAA”, “model risk”, “FinOps” or “agentic”.

AI Risk & Regulatory Readiness

Readiness is decided one use case at a time, not by the AI policy. A launch clears review when someone can show which rules apply to that system, which control answers each one, and where the evidence lives.

Part of Security, Assurance & Governance, in the Think mode. 10 sources, checked .

AI incidents recorded in the AI Incident Database in 2025, up from 233 in 2024
362
AI incidents recorded in the AI Incident Database in 2025, up from 233 in 2024
Stanford Institute for Human-Centered Artificial Intelligence (HAI), 2026(opens in a new tab)
of organizations in IBM's 2026 breach study reported a breach targeting AI models or applications
20%+
of organizations in IBM's 2026 breach study reported a breach targeting AI models or applications
IBM Newsroom (Cost of a Data Breach Report 2026, conducted by Ponemon Institute), 2026(opens in a new tab)
maximum EU AI Act fine for prohibited AI practices, as a share of worldwide annual turnover (or EUR 35m if higher)
7%
maximum EU AI Act fine for prohibited AI practices, as a share of worldwide annual turnover (or EUR 35m if higher)
European Commission, AI Act Service Desk, 2024(opens in a new tab)

The problem

AI use is growing faster than the controls around it. Stanford's 2026 AI Index reports that the AI Incident Database recorded 362 incidents in 2025, up from 233 in 20241. In the same report's survey of business leaders, 62% named security and risk concerns as the main obstacle to scaling agentic AI1. IBM's 2026 breach study found that more than 20% of organizations reported a breach targeting AI models or applications2. Risk teams are right to slow launches down. The harder question is whether anyone can say what would let them say yes.

The rulebook moved in 2026. The EU's Digital Omnibus on AI, Regulation (EU) 2026/1744, was published on 24 July 20263. It moved the AI Act's Annex III high-risk rules to 2 December 2027 and the rules for high-risk AI embedded in regulated products to 2 August 20284. Colorado repealed and replaced its 2024 AI Act with SB 26-189, signed on 14 May 2026, which sets notice, explanation and human-review duties for automated decision-making, with developer documentation due from 1 January 20276. A readiness plan written in 2025 is probably out of date.

Existing frameworks do not cover all of it. The Federal Reserve, OCC and FDIC replaced SR 11-7 with SR 26-2 on 17 April 2026, and its guidance states that generative and agentic AI models are not within its scope7. Banks have to decide for themselves how to govern those systems. Something has to carry them: an AI management system under ISO/IEC 42001, or a program built on the NIST AI Risk Management Framework and its Generative AI Profile8,9.

Good looks like a register of every AI use case with its risk tier, the rules that apply and the control that answers each one. Regulators have said what they expect. NYDFS's October 2024 letter adds no new rules, but it expects Part 500 risk assessments to address your own use of AI and the AI your vendors use10. With that register in place, legal and risk review becomes a check against known requirements rather than an open-ended debate.

What you get

  1. 01

    AI use-case register and risk tiers

    An inventory of the AI you build, buy and switch on inside existing software, each entry tagged with its purpose, data, decision impact and a risk tier you can defend to an examiner.

  2. 02

    Obligation map per use case

    For each use case, the rules that apply and the specific duty each creates: your role and risk category under the EU AI Act, state automated-decision laws such as Colorado's, and sector rules such as SR 26-2, HIPAA or NERC CIP.

  3. 03

    Gap assessment against NIST AI RMF and ISO/IEC 42001

    Where your policies, controls and evidence fall short, scored against the framework your auditors, customers or examiners will use.

  4. 04

    Launch-blocker list and remediation plan

    The specific items standing between each priority use case and sign-off, each with an owner and the evidence that will close it.

  5. 05

    Dated obligation calendar

    The deadlines that apply to you, such as the EU high-risk dates of 2 December 2027 and 2 August 2028, recorded against primary texts so your plan moves when the rules do.

Where it goes wrong

The failure modes we design against from the first week, and the control for each.

  • Treating governance as a policy document

    A policy says what should happen. Reviewers ask what does happen, and without a use-case register and evidence per control, every launch restarts the argument. The control: tie each policy statement to a named control, an owner and an artifact that proves it ran.

  • Assuming model risk management covers generative AI

    SR 26-2 puts generative and agentic AI models outside its scope7. Pushing a chatbot or an agent through the model validation queue either stalls it or approves it against the wrong tests. The control: a separate review path for generative AI with its own testing and monitoring standard.

  • Planning against stale deadlines

    The EU high-risk dates moved under the Digital Omnibus, and Colorado's 2024 law was repealed and replaced3,6. Programs built on 2025 summaries chase dates and duties that no longer exist. The control: an obligation register checked against primary texts every quarter.

  • Missing the AI you did not build

    Vendor features and staff use of public AI tools carry the same data risk as your own models. NYDFS expects risk assessments to cover vendors' AI and suggests blocking staff queries that might expose nonpublic information to a public AI product10. The control: put procurement and SaaS features in the register and assess third-party AI at onboarding.

Standards and rules

  • EU AI Act (Regulation (EU) 2024/1689), as amended by Regulation (EU) 2026/1744

    Primary text on ai-act-service-desk.ec.europa.eu(opens in a new tab)

    If you place AI on the EU market or its output is used in the EU, classify each system now: Annex III high-risk rules apply from 2 December 2027 and product-embedded high-risk rules from 2 August 2028.

  • Colorado SB 26-189 (Automated Decision-Making Technology)

    Primary text on leg.colorado.gov(opens in a new tab)

    Deployers using automated decision-making in consequential decisions must give notice, describe the technology's role in an adverse decision in plain language within 30 days and honor requests for human review; developers owe documentation from 1 January 2027.

  • SR 26-2 / OCC Bulletin 2026-13, Revised Guidance on Model Risk Management (Fed, OCC, FDIC)

    Primary text on federalreserve.gov(opens in a new tab)

    Applies to traditional models and non-generative AI at banking organizations, but excludes generative and agentic AI, so those need their own governance and controls.

  • Sets requirements for establishing, implementing, maintaining and continually improving an AI management system, a common way to show customers and auditors that AI governance runs as a program.

How to measure it

  • Share of AI use cases in the register with a risk tier and a named owner
  • Share of applicable obligations mapped to a tested control with current evidence
  • Time from launch request to legal, risk and security sign-off
  • Number of launches sent back by review for missing evidence
  • Days since the obligation register was last checked against primary texts

Questions buyers ask

Does the EU AI Act apply to a US company?

It can. The Act covers providers that place AI systems on the EU market wherever they are based, deployers located in the EU, and providers and deployers outside the EU when the output of their AI system is used in the EU. A US firm with EU customers, EU staff using its AI, or AI outputs sent into the EU should classify those systems and confirm its role under the Act.

When do the EU AI Act high-risk rules apply after the Digital Omnibus?

Prohibitions have applied since 2 February 2025 and general-purpose AI rules since 2 August 2025. After the 2026 Digital Omnibus, rules for the stand-alone high-risk systems listed in Annex III apply from 2 December 2027, and rules for high-risk AI embedded in products covered by Annex I apply from 2 August 2028. Use the extra time to build evidence, not to pause.

Should we use the NIST AI RMF or ISO/IEC 42001?

Usually both, for different jobs. The NIST AI RMF is a voluntary framework with practical guidance on mapping, measuring and managing AI risk, and its Generative AI Profile adds suggested actions for generative systems. ISO/IEC 42001 sets requirements for an AI management system inside the organization, which suits firms that already run ISO management systems and want a structure auditors can test against.

Sources

Every figure on this page comes from the sources below. We checked each link on .

  1. 1.The 2026 AI Index Report, Chapter 3: Responsible AI(opens in a new tab)Stanford Institute for Human-Centered Artificial Intelligence (HAI), April 2026 · hai.stanford.edu
  2. 2.IBM Study: One in Four Malicious Breaches are AI-Enabled, Costing Companies $6 Million on Average(opens in a new tab)IBM Newsroom (Cost of a Data Breach Report 2026, conducted by Ponemon Institute), July 29, 2026 · newsroom.ibm.com
  3. 3.Digital Omnibus on AI (Legislative Train Schedule)(opens in a new tab)European Parliament, October 25, 2026 · europarl.europa.eu
  4. 4.Timeline for the Implementation of the EU AI Act(opens in a new tab)European Commission, AI Act Service Desk, 2026 · ai-act-service-desk.ec.europa.eu
  5. 5.AI Act (Regulation (EU) 2024/1689), Article 99: Penalties(opens in a new tab)European Commission, AI Act Service Desk, 2024 · ai-act-service-desk.ec.europa.eu
  6. 6.SB26-189 Automated Decision-Making Technology(opens in a new tab)Colorado General Assembly, May 14, 2026 · leg.colorado.gov
  7. 7.SR 26-2: Revised Guidance on Model Risk Management (with attachment SR 26-2a1)(opens in a new tab)Board of Governors of the Federal Reserve System, April 17, 2026 · federalreserve.gov
  8. 8.ISO/IEC 42001:2023 Information technology - Artificial intelligence - Management system(opens in a new tab)ISO and IEC (IEC Webstore listing), December 18, 2023 · webstore.iec.ch
  9. 9.AI Risk Management Framework(opens in a new tab)National Institute of Standards and Technology (NIST), January 26, 2023 · nist.gov
  10. 10.Cybersecurity Risks Arising from Artificial Intelligence and Strategies to Combat Related Risks(opens in a new tab)New York State Department of Financial Services, October 16, 2024 · dfs.ny.gov