AI Vendor Service Level Agreement (SLA)

How to use this document

• What it is for. This is a base contract addendum for agreeing with an AI vendor on service levels (availability, performance, support and credits) and on AI-specific obligations: no training on customer data, notice of model changes, incident notification, control of sub-processors, audits and an orderly exit. The clauses are optional text that your counsel can adapt, combine with the main agreement or use as a list of asks in the negotiation.
• Who should own it. Legal drafts and negotiates; procurement manages the relationship; IT or operations measures the service. Enter the function that manages the contract in [RESPONSIBLE FUNCTION] and the person who owns the vendor relationship in [OWNER].
• Decisions to take first. Before sending it, confirm the vendor's tier under the "AI Vendor Management Policy", which service levels you actually need (not every service justifies 99.9%), which notice periods you will ask for, which positions are non-negotiable and which have a fallback (aid B), and who internally handles each obligation (aid C).
• Fields to complete. Replace the bracketed text, complete Appendices 1 to 3 with the vendor and delete [OPTIONAL] clauses that do not apply. In the text, "Customer" is [LEGAL ENTITY NAME] and "Vendor" is [VENDOR].
• Relationship with other documents. This addendum does not replace the data processing agreement or the security addendum. If those exist, adjust the order of precedence in clause 1.2 and avoid duplicating obligations. Use the "AI Vendor Due Diligence Checklist" first: its findings show which clauses to strengthen. For personal data, use the "AI Data Processing Agreement (DPA)".
• Country variants. This document has no country variants. Check "Related regulatory references" for laws that may require minimum contract content, for example on processors, international transfers, ICT third-party service providers in the financial sector or switching between cloud providers.
• Review. Have counsel review this document before sending it to the vendor and before signing. Numeric values are examples to be adjusted to the vendor's tier and each party's bargaining position.
• Approval and follow-up. Sign the addendum with the main agreement or as an amendment, record its version in "Version history" and schedule the service reviews in aid F.

AI Vendor Service Level Agreement (SLA)

Field	Value
Customer	[LEGAL ENTITY NAME]
Vendor	[VENDOR]
Main agreement	[COMPLETE: name, number and date]
Service covered	[AI SYSTEM]
Effective date	[DATE]
Customer relationship owner	[OWNER]
Vendor relationship owner	[COMPLETE]

1. Purpose and order of precedence

1.1 This addendum sets out the service levels and the artificial intelligence obligations that the Vendor undertakes in providing the Service to the Customer, and forms part of the main agreement.

1.2 In case of conflict, the following prevail, in this order: the data processing agreement, as regards personal data; this addendum, as regards service levels, data use, models and incidents; and the main agreement for everything else. No general terms, usage policy or online terms of the Vendor reduce the obligations in this addendum, even if later accepted electronically.

2. Definitions

• Service: the service described in Appendix 1, including its AI features, application programming interfaces and support.
• Customer Data: all information that the Customer or its users enter, upload or make available to the Service, including instructions, messages, documents, files, fine-tuning data, indexes and memories created from them, Outputs and the associated logs.
• Outputs: the content, predictions, scores, classifications, recommendations or actions the Service generates for the Customer.
• Model: any AI model the Service uses, whether the Vendor's own or a third party's.
• Material Model Change: replacing the Model, a version change, retraining, a change of model provider or a configuration change that may noticeably alter the quality, behavior, safety or characteristics of Outputs, or the processing of Customer Data.
• Security Incident: any unauthorized access to, or use, disclosure, alteration, loss or destruction of, Customer Data, or any compromise of the systems that process it.
• AI Incident: behavior of the Service that systematically produces erroneous, biased, harmful or disallowed Outputs, or unauthorized actions, that affects or may affect the Customer or third parties, including manipulation of the Model.
• Sub-processor: any third party that processes Customer Data on the Vendor's behalf, including model and infrastructure providers.
• Availability: the percentage of time in a calendar month during which the Service is operational and responds in line with its documentation.
• Service Credit: the discount on fees the Vendor applies when it misses a service level.