AI Services Agreement

An AI services agreement governs hosted or managed AI functionality, inputs, outputs, model behavior, data use, evaluation, and human oversight.

A useful document turns the parties' actual arrangement into measurable duties, approvals, timing, remedies, and a reliable execution record. Its terms should be reconciled to the transaction rather than copied from an unrelated form.

• A business buys a hosted generative assistant
• A vendor automates document classification for a customer
• An enterprise deploys an AI workflow using confidential prompts

The document name alone does not determine the right structure. Consider a different instrument or qualified legal review when any of these conditions applies:

• Do not assume generated outputs are accurate, unique, or eligible for copyright protection.
• Do not use it alone for a high-impact regulated decision without domain-specific review and controls.

Record exact facts before clauses are written. Names, authority, dates, amounts, defined terms, dependencies, and incorporated materials should be verifiable and consistent.

• Models, versions, use cases and excluded decisions
• Inputs, personal data, training and retention settings
• Evaluation metrics, human review and escalation
• Output rights, third-party claims, pricing and termination

These decisions shape the allocation of responsibility and should not be left for boilerplate to decide:

• Can customer data improve shared models
• Who validates output before use
• How material model changes are communicated
• What records support audit and dispute resolution

• Defined AI services, approved use and model-change notice
• Input rights, data isolation, retention and training choices
• Output use rights, provenance and third-party restrictions
• Evaluation, monitoring, safety controls and human oversight
• Incidents, suspension, export and deletion

Every provision should use the same parties, dates, standards, defined terms, and document hierarchy. A clause that is reasonable by itself can still create a conflict when it is not reconciled with payment, default, termination, or another exhibit.

1. 01Describe the intended result and the relationship in plain language.
2. 02Confirm parties, authority, governing jurisdiction, dates, money, property, services, and approvals.
3. 03Resolve the key decisions and identify every schedule, exhibit, disclosure, consent, or filing.
4. 04Draft the provisions as one consistent system, then review the complete execution set before signature.

Confidential inputs may be reused beyond expectations; output errors, bias, infringement, and model changes can undermine a promised workflow.

The main agreement may establish the framework while schedules, exhibits, disclosures, consents, or operational records supply transaction-specific details.

• Model and feature inventory
• Data flow and evaluation plan
• Security and processing addendum

Each incorporated document should be identified precisely, use the same names and effective date, and follow a stated order of precedence if terms conflict.

Approve use cases and input classes, test on representative data, set human-review checkpoints, and document model versions and data settings.

• Confirm legal names, roles, capacity, addresses, and signing authority
• Reconcile dates, amounts, definitions, cross-references, schedules, and exhibits
• Confirm that duties, deadlines, approvals, acceptance standards, and payment triggers are measurable
• Check that default, termination, remedies, and surviving obligations work together
• Complete jurisdiction-specific forms, notices, witnesses, notarization, filings, or professional review when applicable
• Deliver and preserve the complete signed package with its incorporated documents

These sources provide federal, state-resource, regulatory, or institutional context. They do not replace checking the law and required forms applicable to the parties, transaction, and governing jurisdiction.

1. Source 1

   AI Risk Management Framework

   National Institute of Standards and Technology. Official voluntary framework addressing trustworthy AI design, measurement and management.

2. Source 2

   Keep your AI claims in check

   Federal Trade Commission. Official FTC guidance on substantiating claims about AI products.

3. Source 3

   AI Companies: Uphold Your Privacy and Confidentiality Commitments

   Federal Trade Commission. Official FTC analysis of stated privacy commitments and AI training data uses.