How to Write AI Terms into a Vendor Contract

Current as of October 30, 2026. This guide draws on the EU Digital Operational Resilience Act (DORA) and AI Act, SEC Regulation S-P, OSFI guidelines, New York and Illinois insurance guidance, EIOPA and PRA supervisory texts, FINRA's 2026 oversight report, a proposal from the US banking agencies and model AI contract clauses supported by the European Commission, as published on that date. Which of them apply depends on the firm; it is not legal advice.

Our guide to AI vendor due diligence covers finding, questioning and testing vendor AI. This guide turns what diligence found into enforceable clauses.

Unless a step quotes a rule, it is a recommendation, labeled (interpretation). No text reviewed supplies clause wording for AI vendor contracts in financial services; the drafting suggestions below are ours.

Step 1: Identify which contract rules govern this AI arrangement

DORA applies to credit institutions, investment firms, insurance and reinsurance undertakings and other listed EU financial entities, with some smaller entities excluded. It defines ICT services as "digital and data services provided through ICT systems to one or more internal or external users on an ongoing basis"; an AI tool delivered as a hosted service would ordinarily fit (interpretation).

In Canada, B-10 sets out OSFI's expectations and is "applicable to all federally regulated financial institutions, including foreign bank branches and foreign insurance company branches" (branches had until March 31, 2025): "OSFI expects the FRFI to include in written agreements for high-risk and critical arrangements the provisions that are set out in Annex 2 of this Guideline." Standardized contracts and arrangements with no written contract are handled under its Section 3 instead. Arrangements that predate May 1, 2024 were to be "reviewed and updated at the earliest opportunity". OSFI's model risk guideline E-23, effective May 1, 2027, covers models sourced from third-party vendors "pursuant to" B-10. Regulation S-P covers broker-dealers, funding portals, investment companies, SEC-registered advisers and registered transfer agents. New York's Circular Letter No. 7 (July 11, 2024) covers insurers' AI in underwriting and pricing; Illinois Bulletin 2024-08 (March 13, 2024) covers insurers' AI systems used in regulated insurance practices. Both state supervisory expectations, not statutory contract requirements.

Record which of these texts applies to each AI contract (interpretation).

Step 2: Define the AI service, its intended use and the model behind it

DORA requires "a clear and complete description of all functions and ICT services to be provided by the ICT third-party service provider".

For EU insurers, EIOPA's August 2025 opinion, addressed to national supervisors and covering AI systems that are neither prohibited nor high-risk under the AI Act, says that where third-party providers' intellectual property rights make some governance measures, such as data governance or explainability, hard to implement, undertakings should adopt complementary governance measures and other measures "such as including appropriate clauses in contracts and service level agreements, conducting external audits, or performing due diligence testing and monitoring."

Name in the service description the intended use, prohibited uses, underlying models and versions, required documentation and tested performance measures (interpretation).

Step 3: Limit how the vendor may use the firm's data

DORA requires the locations where data is processed, "including the storage location", with advance notice of changes, and provisions on "availability, authenticity, integrity and confidentiality" of data. B-10 Annex 2 says the agreement "should also specify whether and how the third party has the right to use the FRFI's assets", data included. FINRA's 2026 oversight report for its member firms lists, among effective practices, "adding language that prohibits firm or customer sensitive information from being ingested into a third-party vendor's open-source GenAI tool"; FINRA says the report creates no new requirements.

Bar training of models for others on the firm's inputs, outputs or customer data without written consent, and settle ownership of outputs and fine-tuned models (interpretation).

Step 4: Require notice and control of model and subcontractor changes

DORA requires "notification of any development that might have a material impact on the ICT third-party service provider's ability to effectively provide the ICT services supporting critical or important functions in line with agreed service levels". B-10 Annex 2 says the agreement should "require the third-party to notify the FRFI of any subcontracting of services".

Under the EU AI Act, a deployer that makes "a substantial modification to a high-risk AI system" that remains high-risk becomes its provider. Article 25(4), as amended by Regulation (EU) 2026/1744, says a high-risk system's provider and a third party supplying "an AI system, AI model, tools, services, components, or processes" used in it "shall, by written agreement, specify the necessary information, capabilities, technical access and other assistance based on the generally acknowledged state of the art, in order to enable the provider of the high-risk AI system to fully comply with the obligations set out in this Regulation." For Annex III systems, including creditworthiness evaluation or credit scoring of individuals and life and health insurance risk assessment and pricing, these obligations apply from December 2, 2027. It binds a firm only as a provider: a firm that builds its own high-risk system, such as a credit-scoring model, on a vendor's model or tools would need such an agreement with that vendor, while a firm that only deploys a vendor's system is not a party to it (interpretation). The paragraph does not apply to third parties that make tools, services, processes or components, other than general-purpose AI models, publicly available under a free and open-source license.

Require advance notice of new underlying models, model providers or AI features, with a right to re-test or exit (interpretation).

Step 5: Secure the information needed to validate and explain outputs

For UK banks, building societies and PRA-designated investment firms with internal model approval (not insurers), PRA SS1/23 expects firms to "satisfy themselves that the vendor models have been validated to the same standards as their own internal MRM expectations". New York's circular letter says an insurer "may not rely on the proprietary nature of a third-party vendor's algorithmic processes to justify the lack of specificity related to an adverse underwriting or pricing action."

Oblige the vendor to provide validation evidence and the information needed to explain an individual decision, under confidentiality terms if necessary (interpretation).

Step 6: Write audit, information and regulator-access rights

For critical or important functions, DORA requires "unrestricted rights of access, inspection and audit by the financial entity, or an appointed third party, and by the competent authority", and for all ICT contracts a duty "to fully cooperate with the competent authorities and the resolution authorities of the financial entity". B-10 says the firm "should also have the right to conduct or commission an independent audit of a third party." New York and Illinois point insurers, where appropriate and available, to terms providing audit rights or audit reports and regulatory cooperation; Illinois's wording is that the third party "cooperate with the Insurer with regard to regulatory inquiries and investigations".

Extend audit scope to AI controls: training-data provenance, bias and accuracy testing, change logs and incident records (interpretation).

Step 7: Set incident, breach notice and cooperation terms

Regulation S-P requires covered institutions' policies to be reasonably designed to ensure service providers "take appropriate measures to" notify the institution "as soon as possible, but no later than 72 hours after becoming aware that a breach in security has occurred resulting in unauthorized access to a customer information system maintained by the service provider." A covered institution "may enter into a written agreement with its service provider to notify affected individuals on the covered institution's behalf", but the obligation to ensure they are notified "rests with the covered institution." DORA requires the provider "to provide assistance to the financial entity at no additional cost, or at a cost that is determined ex-ante" when an ICT incident related to the service occurs.

Define AI incidents too, such as materially wrong or biased outputs, with notice periods (interpretation).

Step 8: Plan exit, data return and records

For critical or important functions, DORA requires exit strategies with "a mandatory adequate transition period", and for all ICT contracts return of data "in an easily accessible format" on insolvency, discontinuation or termination. B-10 says data and records "should be returned to the FRFI in a format that allows the FRFI to sustain business operations without unreasonable expense." For broker-dealers, FINRA's 2026 report lists, among its recordkeeping findings, "not confirming that service contracts and agreements comply with applicable recordkeeping requirements, including records stored by third-party vendors."

Under the AI Act, from December 2, 2027 for Annex III systems, deployers of high-risk systems must keep automatically generated logs under their control for "at least six months" unless other law provides otherwise; financial institutions keep them within their financial services documentation. Make the vendor preserve and export logs, inputs and outputs the firm must retain (interpretation).

When the firm cannot negotiate

The US banking proposal says a bank with limited negotiating power "may still reasonably proceed with the relationship" in some circumstances. DORA says parties "shall consider the use of standard contractual clauses developed by public authorities for specific services." The model contractual clauses for public procurement of AI (version February 2025) were written for public buyers; their disclaimer calls them a working document that does not reflect an official Commission position, and the commentary says it would not be appropriate for organizations other than public organizations to use them, though parts could be used by other organizations, which "will always have to assess, clause-by-clause, whether the use of the MCC-AI is appropriate in each situation." Neither the clauses nor the commentary mentions DORA or financial services, so they are not a substitute for Article 30; parts can serve as a drafting reference (interpretation).

Checklist

Items restate the recommendations above (interpretation).

• The texts that apply are recorded
• The service description names use, prohibited uses, models and documentation
• Data use, training, locations and output ownership are settled
• Model and subcontractor changes require notice and allow re-testing or exit
• The vendor supports validation and explanation
• Audit and regulator-access rights cover AI controls
• Breach notice meets Regulation S-P where it applies; AI incidents are defined
• Exit, data return and log terms match records obligations
• Accepted gaps in standard contracts are documented
• The proposed US interagency guidance is tracked until finalized

How we sourced this guide: every factual statement in it is drawn from published rules, regulatory guidance and other public documents, and we keep a full set of sources and citations for it. If you would like them, or want to question a point, get in touch through our contact page and we will share them.