Book a demo
Topics · AI-DLC

Payments AI-DLC SR 26-2 | Moring

SR 26-2, the revised interagency model risk guidance issued April 17, 2026 by the Federal Reserve, OCC, and FDIC, shifts US supervision to a principles-based framework with materiality provisions. Its footnote excludes generative and agentic AI from scope, so payments institutions must re-paper regulated models against the new text while governing agent-driven development through their own documented programs.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

SR 26-2, the revised interagency model risk guidance issued April 17, 2026 by the Federal Reserve, OCC, and FDIC, shifts US supervision to a principles-based framework with materiality provisions. Its footnote excludes generative and agentic AI from scope, so payments institutions must re-paper regulated models against the new text while governing agent-driven development through their own documented programs.

How it works

SR 26-2 is the revised interagency guidance on model risk management issued on April 17, 2026 by the Federal Reserve Board, the OCC, and the FDIC. It is the first rewrite of US model risk supervision in fifteen years, and it replaces the 2011-era framework with a principles-based approach built around core model risk principles and new materiality provisions rather than prescriptive requirements.

For payments organizations, the guidance matters in two directions at once. First, payment businesses that operate models for fraud scoring, authorization, credit-adjacent decisions, or risk pricing fall squarely inside the model risk tradition, and an existing program needs re-papering against the new text rather than retirement. Practitioner analysis stresses that SR 26-2 is a calibration, not an overhaul: institutions with mature programs reassess validation priorities under the materiality provisions instead of rebuilding governance from scratch.

The second direction is the scope boundary. A footnote in the guidance explicitly places generative and agentic AI outside its scope, which means the LLM applications and agents spreading through payments engineering and payment operations have no supervisory blueprint. Industry commentary is blunt about the consequence: SR 26-2 regulates your models, not your AI agents, so the control decision for agentic systems moves to the institution's own governance program.

That boundary creates a concrete design question for payments teams adopting agentic AI. Model purpose and model exposure drive materiality under the revised text, and similar logic applies by extension to agent work: an agent that only drafts internal documentation sits in a different class from one that touches settlement parameters or dispute workflows. Institutions that map agent use cases onto a consequence hierarchy can align their treatment of agentic risk with the spirit of the model risk framework even where the letter does not reach.

Examiners are not waiting for the next rewrite. Analysis of the generative AI carve-out notes that examiners still test how institutions govern AI outside the rulebook, which means the practical defense is documented governance: an inventory of agentic systems, risk assessment by consequence, human oversight for consequential actions, and evidence that controls operated. For payments, where operations already sit under intense supervisory attention, treating the carve-out as a free pass would be a serious misreading of the environment.

The guidance also carries organizational implications. Analysis of the transformation describes the MRM function evolving from a compliance exercise toward a strategic capability, and payments institutions can use the same shift: validation and independent review capacity becomes the resource that both regulated models and, by internal policy, agentic systems draw on. Aligning agent governance with the model risk operating model, rather than building a parallel structure, avoids duplicated effort and inconsistent standards.

The practical takeaway for payments leadership is sequencing. Re-paper existing regulated models against SR 26-2 first, since that is where supervisory expectations are explicit; then formalize internal governance for agentic AI, using the guidance's principles of purpose, exposure, materiality, and effective challenge as the template. Institutions that do both will find the carve-out manageable; those that do only the first will carry ungoverned agent risk into an environment where examiners are already asking questions.

Frequently asked questions

What is SR 26-2?

SR 26-2 is the revised interagency supervisory guidance on model risk management issued on April 17, 2026 by the Federal Reserve Board, the OCC, and the FDIC, replacing the 2011-era framework after fifteen years. It shifts from prescriptive requirements to a principles-based approach with core model risk principles and materiality provisions. A footnote explicitly excludes generative and agentic AI from its scope, so institutions must govern those systems through their own programs while applying the guidance's logic of purpose, exposure, and effective challenge.

Does SR 26-2 apply to agentic AI in payments?

Not directly. The guidance expressly places generative and agentic systems outside its scope, so agents used in payments engineering or operations have no supervisory blueprint to follow. The consequence is that control responsibility rests with the institution: risk tiering by consequence, human oversight of consequential actions, documented testing, and retained evidence. Examiners still test governance of AI outside the rulebook, so payments organizations should formalize internal agent governance aligned with model risk principles rather than treating the carve-out as an exemption from oversight.

What should payments institutions re-paper under SR 26-2?

Existing regulated models, including fraud scoring, authorization, and credit-adjacent models common in payments, should be re-papered against the new text rather than retired, because the guidance supersedes prior model risk requirements. Practitioner analysis describes the change as a calibration for mature programs: reassess validation priorities under the new materiality provisions and confirm the program reflects the core principles. Sequencing matters, so re-papering regulated models comes first while internal governance for agentic systems, which sits outside the guidance, is formalized in parallel.

How does the materiality concept affect payments model governance?

Materiality provisions let institutions concentrate validation and independent review where model exposure is greatest, replacing uniform treatment with risk-proportionate effort. A payments organization can apply the same logic to its growing agent estate: an agent drafting internal documentation is materially different from one touching settlement parameters or dispute workflows, and oversight should differ accordingly. Mapping agent use cases onto a consequence hierarchy aligned with materiality thinking keeps scarce validation capacity pointed at the systems where failure would actually harm the business or its customers.

Sources

  1. www.federalreserve.gov
  2. www.deloitte.com
  3. kpmg.com
  4. domino.ai
  5. www.ey.com
  6. ntdciso.com
  7. cimcon.com
  8. www.fiddler.ai
  9. - federalreserve.gov
  10. - ithandbook.ffiec.gov
  11. - artificialintelligenceact.eu
  12. - eur-lex.europa.eu