Back to Insights
Essay

October 22, 2026

Fannie Mae SI vs. Freddie Mac AI/ML: A Comparison

By Suha Beidas Zehl, CMB®, AMSP™

Cover image for the article Fannie Mae SI vs. Freddie Mac AI/ML: A Comparison

At a glance

As of January 18, 2027, both GSEs will embed AI governance inside their information security frameworks. Fannie Mae's new Section 4 is now more detailed than Freddie Mac's Section 1302.8 on the AI lifecycle, while Freddie Mac remains stronger on audits, executive sign-off, and indemnification.

Fannie MaeFreddie Mac
SourceInformation Security and Business Resiliency Supplement, Section 4, Super Intelligence (SI)Single-Family Seller/Servicer Guide Section 1302.8, Use of artificial intelligence and machine learning
Issued bySupplement Bulletin 26-01, published October 20, 2026Guide Bulletin 2025-16, issued December 3, 2025
EffectiveJanuary 18, 2027March 3, 2026 (in force now)
PredecessorInformation Security and Business Resiliency Supplement (August 12, 2025 edition); and for AI provisions, Lender Letter LL-2026-04 (effective August 6, 2026)Earlier 1302.8 (applicable law and indemnification only)
Who it coversEvery "Company" under the Supplement: Single-Family lenders, Multifamily lenders, document custodians, and integrators (third-party solution providers)Single-Family Seller/Servicers
Term used"Super Intelligence," formally defined; Fannie Mae says the name change does not alter scope"AI/ML"; not formally defined in the Guide
TriggerUse of an SI system in connection with activities under, or relating to, the Underlying AgreementsUse of AI/ML in connection with originating or servicing Freddie Mac mortgages

Side-by-side: Section 4 vs. Section 1302.8

The two frameworks share the same core: a documented program, risk management to the company's tolerance, vendor oversight, and disclosure on request. They diverge on lifecycle detail, audits, executive sign-off, and liability.

RequirementFannie Mae Section 4Freddie Mac Section 1302.8Goes further
Governance programDocumented and implemented formal SI governance programClear governance frameworks for mapping, measuring, and managing AI risk, with trustworthy AI characteristics built into policies and practicesEven
Executive accountabilityA designated owner implements, maintains, and reviews policies; the Supplement separately requires a senior executive for the overall security program and an annual officer attestationPolicies approved by Senior Management, at minimum the CIO, CTO, CISO, or CRO (or equivalents)Freddie Mac
Annual reviewDesignated owner reviews at least annually for compliance and industry best practiceOwner(s) implement, maintain, and review policies at least annually for compliance with Applicable Law and industry best practicesEven
Risk toleranceRisk management activities based on the Company's risk toleranceProcesses to set the level of risk management needed based on risk toleranceEven
AI use case inventoryRequired: inventory of SI use cases, with processes to detect, track, and measure purpose and scopeNot stated as a standalone requirement; implied by the disclosure-on-request obligationFannie Mae
Data governanceData governance, retention, classification, and quality standardsCovered through risk management and data protection expectationsFannie Mae
Model lifecycleGoverns model design, training, testing, and validation; change management for models, systems, and use cases; post-deployment monitoring and retirementRegular monitoring of AI systems for performance, security breaches, and biasFannie Mae
AuditsMonitoring, auditing, and ongoing oversight required, without specifying internal and external audits; general annual independent security assessment applies (Section 3.3)Regular internal and external audits to find vulnerabilities or policy deviations; planned ongoing monitoring and periodic review, with defined roles and review frequencyFreddie Mac
Framework alignmentOverall security program must meet or exceed NIST 800-53 or ISO 27001 (Section 3); not SI-specificAudits to confirm compliance with standards like NIST 800-53 and ISO 27001, tied to the AI frameworkFreddie Mac
AI security threatsAssess adversarial manipulation, model inversion, data leakage, prompt injection, and training data integrity; document safeguards for data protection, model integrity, and misuseAssess AI/ML for specific threats such as data poisoning and adversarial inputs; monitor for security breaches; Section 1302.2 annual training should incorporate AI-powered tactics (deepfakes, targeted phishing) and threats to AI systems (model inversion, data poisoning, prompt injection), plus AI risk management training when necessaryEven (both assess; Fannie names more threats, Freddie adds training)
AI incident playbooksNo AI-specific incident playbook requirementSection 1302.2(b)(xvi) requires documented remediation action plans, including playbooks for AI/ML incidentsFreddie Mac
Segregation of dutiesRequired, to prevent conflicts of interest among those mapping, measuring, and managing SI riskRequired, to prevent conflicts of interestEven
Communication to staffTransparent communication with personnel whose jobs involve SIFramework communicated to personnel in areas that use AI/MLEven
Ethics and lawTrustworthy, ethical, and responsible SI principles; understanding of and compliance with applicable lawCompliance with Applicable Law and Purchase Documents; legal requirements understood, managed, and documentedEven
Confidential data and outputsControls and training to protect Confidential Information when using SI; SI outputs derived from Confidential Information are Confidential InformationCovered by general Chapter 1302 confidentiality and data protection rulesFannie Mae
Third partiesThird-party SI governance no less protective; Section 3.14 makes the Company responsible for third-party failures as if its ownNo vendor clause in Section 1302.8 itself; vendor AI/ML is addressed in Freddie Mac's FAQ and reached through the Seller/Servicer under Chapter 1302 Related Third Party rulesFannie Mae (explicit liability for third-party failures)
Disclosure on requestTypes of SI, purpose and manner of use, safeguards, and anything else Fannie Mae requiresTypes of AI/ML, purpose and manner of use, safeguards, and anything else Freddie Mac requiresEven (near identical)
IndemnificationNone specific to SIBroad indemnity for all losses arising directly or indirectly from AI/ML use, including indirect and consequential damagesFreddie Mac

Incident notification windows

Both GSEs now use a 36-hour window. Freddie Mac adds an immediate-notice trigger and a quarterly path for small privacy incidents; Fannie Mae has neither.

Freddie Mac’s reporting time frames now sit in Section 1302.5, after a Chapter 1302 refactor that cut the window from 48 to 36 hours, effective January 1, 2025 (Bulletin 2024-17). Freddie Mac restated these time frames in Bulletin 2026-8.

Fannie Mae (Supplement Section 5)Freddie Mac (Guide Section 1302.5)Goes further
Standard windowWithout undue delay, no later than 36 hours after identifying an incident or reasonably concluding one may have occurredNo later than 36 hours after discoveryEven
Immediate noticeNo separate trigger; Section 3.6 requires prompt notice of disruptions that could affect business with Fannie MaeImmediately if an Incident causes the Seller/Servicer to shut down, disable, or stop using any system it ordinarily uses to originate or service Freddie Mac mortgages, including third-party systemsFreddie Mac
Small privacy incidentsNo exception; every Cybersecurity Incident (defined to include data breaches) follows the 36-hour rule, regardless of how many borrowers are affectedNon-critical Privacy Incidents (fewer than 10 Freddie Mac borrowers, none of the listed aggravating factors) reported quarterly, by the 5th of January, April, July, and OctoberFannie Mae (no carve-out)
Follow-upPrompt supplemental details as they become availableInterim status updates until Freddie Mac is satisfied the Incident is resolved and closed, plus a final closure report and a Certificate of ComplianceFreddie Mac
Information requiredInitial notice: identification date, containment status, and any suspended operations. Then, when available: indicators of compromise, attack methodology, IR vendors, law enforcement, investigation leads, and Fannie Mae data exposed (Section 5). Section 5.3 adds number of individuals impacted, jurisdictions, data attributes, and root cause.Initial notice: investigation lead's contact details, law enforcement involved, and all known facts. Then: technical and forensic details, remediation and recovery steps, postmortems, data exposed and likelihood of misuse, whether cyber or other insurance covers the incident, potential damage estimates, and compliance actions (Section 1302.5(a)).Freddie Mac (broader list)
Access restoration and closureRestored at Fannie Mae's sole discretion after an attestation as to the security and safety of the Company's systems, signed by either an independent third-party security professional or two Company officers, plus any other conditions Fannie Mae imposes (Section 5.1). No formal closure report or certificate required.Restored at Freddie Mac's sole discretion, conditioned on any assurances it requests plus an attestation by a duly authorized corporate officer covering containment, eradication, and remediation, including removal of the threat actor (Section 1302.5(c)). Closure also requires interim updates, a final incident closure report, and a Certificate of Compliance with breach notification laws and the Guide (Section 1302.5(a)).Freddie Mac (Fannie Mae more flexible on who attests)
Regulatory filingsCopy or link to any SEC or regulatory incident filing within 36 hours of filing (Section 5.2); copy of the notification letter sent to affected individuals; permission from Fannie Mae's Privacy Office before naming Fannie Mae in any borrower or regulator notice (Section 5.3)Copies of any communications to Borrowers, regulators, State and federal agencies, credit reporting agencies, or others, provided as part of ongoing updates; no SEC-specific clause or fixed deadline. Freddie Mac must have the opportunity to review and comment on any borrower notice that refers to it before it goes outSplit: Fannie Mae on filing deadlines, Freddie Mac on correspondence oversight
Physical informationLost, stolen, or misrouted paper or media: same 36-hour rule (Section 5.4)No separate rule; covered by the general Incident definition and its 36-hour windowEven
Where to reportprivacy_office@fanniemae.comFreddie Mac's online incident portalNot applicable

On external notices, Fannie Mae is stricter on filing deadlines, while Freddie Mac goes further on oversight of agency and borrower correspondence. Freddie Mac also asks for insurance coverage and damage estimates, which Fannie Mae does not.

On closure, Freddie Mac goes further: a defined closure report, a Certificate of Compliance, and an officer attestation tied to specific eradication steps. Fannie Mae's attestation is broader but lets a company use an independent third party instead of officers. Incident response playbooks for dual-GSE lenders should plan for both.

Neither GSE addresses law enforcement requests to delay notification. Many state breach laws allow such delays, so incident response teams should track any request and its timing separately, as a state-law matter rather than a GSE requirement.

For AI specifically: neither GSE sets a separate AI incident clock. An AI-related event that meets either definition of an incident, such as data leakage through a model or a prompt injection that exposes borrower data, follows the standard window. Freddie Mac does require AI/ML-specific incident playbooks (Section 1302.2(b)(xvi)).

What this means for lenders and their technology partners

The old advice to "build to Freddie Mac and you satisfy both" no longer holds. Each GSE now has requirements the other lacks, so a dual-GSE program needs the stricter element of each.

Fannie Mae now goes further on:

A formal AI use case inventory

Model lifecycle controls: design, training, testing, validation, change management, and retirement

Named AI security threat assessments, including prompt injection and model inversion

Treating AI outputs derived from Confidential Information as Confidential Information

Explicit liability for third-party failures, and reach beyond Single-Family to Multifamily lenders, document custodians, and integrators

Freddie Mac still goes further on:

Regular internal and external audits of the AI framework against NIST 800-53 or ISO 27001

Senior management approval of AI policies

A broad AI-specific indemnification, including consequential damages

AI-specific threat training under Section 1302.2

An immediate-notice trigger for incidents that force system shutdowns

Documented incident response playbooks for AI/ML incidents

Practical steps before January 18, 2027:

Build or update the AI use case inventory, including AI embedded in vendor platforms.

Add lifecycle and retirement controls to existing model risk or AI policies.

Map named AI threats to existing security testing and controls.

Update data classification so AI outputs built from borrower or Fannie Mae data are handled as Confidential Information.

Revisit third-party contracts to flow down SI controls, given liability for vendor failures.

Confirm technology vendors and document custodians know the new SI section applies to them as Companies under the Supplement.

Caveats

Fannie Mae details come from the full Supplement text and Bulletin 26-01.

Sections 1302.2, 1302.5, and 1302.8 are checked against the Guide text (effective July 1, 2026, January 1, 2025, and March 3, 2026, respectively).

Section 1302.8 has no third-party clause; how vendor AI is governed rests on Freddie Mac's FAQ and broader Chapter 1302 rules.

Freddie Mac's Multifamily Guide may carry parallel AI provisions; it is not compared here.

Sources

Fannie Mae Information Security and Business Resiliency Supplement (effective Jan. 18, 2027)

Fannie Mae Supplement landing page and FAQ

Freddie Mac Guide Section 1302.8

Freddie Mac Guide Section 1302.5

Freddie Mac Guide Section 1302.2

Freddie Mac Incident Reporting FAQ

Freddie Mac Bulletin 2024-17

Freddie Mac Bulletin 2026-8

National Mortgage Professional: The Algorithmic Servicer

MSLaw: Freddie Mac Guide Bulletin summary

MQMR: Freddie Mac AI Governance FAQ

Pease Bell: Fannie Mae AI Governance

First Line Compliance: Freddie Mac AI Governance

Enjoyed this one? There’s plenty more where it came from.

Browse all articles →