Key Takeaways
- •Weights, code, data and documentation can have different owners and licenses. Identify each included component and its actual permissions.
- •Use, fine-tuning, hosting and redistribution are separate grants. Match them to scope, fees, duration and downstream conditions.
- •Fine-tuning does not automatically establish a copyright derivative or ownership of all resulting weights. Private promises do not create missing output copyright.
- •This custom form does not incorporate or certify Apache, GPL, RAIL or another standard license. Open weights, open source and regulatory exemptions require separate analysis.
- •EU applicability, duties and exceptions depend on actual facts, roles and dates. The builder records that assessment and keeps unresolved questions as an unsigned draft.
Reviewed for accuracy by the document.com legal team. Educational information, not legal advice.
What Is AI Model License Agreement?
An AI model license agreement records permission to use identified model components and related commercial obligations. It distinguishes actual copyright and other rights from file access, contractual promises and regulatory responsibilities.
The builder supports custom permissive, use-restricted, reciprocal-source and proprietary proposals. These labels organize the intended approach; they do not substitute for the complete terms of a standard license or certify an open-source release.
Four model-license examples
These are unsigned custom drafts, not standard public-license texts. Choose actual component rights and terms in the builder.
Why This Matters Now
A release can mix separately licensed code, weights, tokenizer files, data and documentation. A general statement that a model is open does not resolve every component’s terms.
Fine-tuning, hosted access, weight distribution and output use raise different permission and responsibility questions. The agreement should make those choices explicit.
New deployment locations, modifications or use cases can change applicable duties. Record actual roles, evidence, dates and change procedures instead of relying on a generic compliance statement.
The Legal Backbone
Actual rights and creator chain
Copyright definitions, protected expression, ownership and transfer requirements apply to the actual components and facts. A fine-tuned model is not automatically a copyright derivative, and an assignment cannot convey rights a party does not hold.
Open-source claims require actual license terms
The OSI definition concerns distribution terms and includes criteria beyond access to code. A use restriction or a source-available label does not by itself establish open-source status. This custom agreement does not incorporate a standard license by analogy.
Output ownership remains limited
The Copyright Office’s 2025 report distinguishes generated material from qualifying human expression. Private promises can allocate lawful duties and actual held rights but cannot create copyright, eliminate third-party claims or bind strangers.
EU Article 53 and its limited exception
Actual GPAI duties include relevant documentation, copyright policy and a public training-content summary. The limited free/open-source documentation exception requires the statutory conditions and does not remove the copyright-policy or summary duties.
Systemic-risk and transition rules
Article 55 adds duties for providers of models with systemic risk, including evaluation, mitigation, serious-incident reporting and cybersecurity. Determine actual classification and applicable dates, including transition rules; a contract selection is not certification.
Risk and ending terms need consistency
Specify real warranties, indemnity, payment conditions, notice/cure and surviving rights. Preserve lawful records and legal holds, and assess existing downstream grants separately.
Define component rights and each permitted activity
List exact model/version identifiers, included files, exclusions, owners and upstream license versions. Check actual compatibility and intended-use permissions.
Define users, purposes, territory, capacity and actual rights licensed. Decide fine-tuning, hosting and redistribution independently.
For distribution or reciprocal-source terms, identify recipient rights, notices, acceptance, release materials, triggers and delivery methods. Do not substitute a broad license label.
Record actual modifications/output rights and any explicit scoped assignment. Avoid automatic ownership of all future weights, outputs or unrelated code/data.
Match execution or payment timing to actual fees, royalties and duration. State real termination, cure, refunds, surviving downstream uses and lawful retention.
Allocate actual compliance and operational tasks without treating the agreement as regulatory approval. Resolve rights, applicability and review before preparing signature fields.
When You Need This
For a custom agreement covering identified model components and actual rights between known parties.
For a negotiated model-use arrangement where permissions, compensation, distribution and responsibilities need to be recorded separately.
If adopting a standard public license, use and assess its actual complete terms and component compatibility. This builder is not a replacement text or certification for that license.
How to Fill Out AI Model License Agreement
1. Identify the parties and model
Record authority, locations, release identifiers and included/excluded components.
2. Check upstream rights
Identify exact license versions, notices, compatibility and permissions for actual intended uses.
3. Choose separate permissions
Define use, tuning, hosting and distribution, with real recipient and source-delivery terms.
4. Set actual rights and fees
Specify output/modification arrangements, any scoped assignment, payment trigger and actual commercial terms.
5. Resolve compliance and ending terms
Record actual roles, duties, evidence, dates, risk allocation, notice/cure and lawful retention.
6. Review before execution
Reconcile schedules, upstream terms and prior agreements. Keep unresolved rights or applicability as an unsigned draft.
Key Terms Defined
- Included components
- The identified weights, code, tokenizer, data or documentation actually covered by the grant.
- Fine-tuning permission
- A defined right to modify specified components, distinct from ownership or a legal determination of derivative-work status.
- Redistribution
- Conveying identified components to recipients under actual rights, notices and other lawful conditions.
- Reciprocal-source obligation
- A specific duty to provide identified materials when an agreed activity occurs; a label alone does not define it.
- Regulatory role
- A responsibility determined by actual conduct and applicable law; private task allocation does not erase it.
Related Documents
Custom model license
Records actual component permissions, modification/output arrangements, fees, compliance tasks and ending terms.
Standard public license
Uses its own complete, specified terms. A comparison or checkbox does not incorporate or certify those terms.
Hosted API terms
Address a service relationship and API access; they do not necessarily convey model weights or redistribution rights.
Legal Authorities & Sources
This page is grounded in primary law. The statutes and official resources below are the authorities behind the guidance above. Verify the current text of any statute before relying on it.
- Copyright Act: definitions and scope
- Copyright Act: ownership and transfers
- Copyright Office: AI copyrightability report (2025)
- EU Commission: Article 53
- EU Commission: Article 55
- EU Commission: Article 111 transition rules
- EU Commission: general-purpose-model questions and answers
- Open Source Initiative: Open Source Definition
- Protected reporting, 18 USC 1833
Frequently Asked Questions
Create your AI Model License Agreement in minutes.
Answer a few questions and download a clear, attorney-drafted document that cites the controlling law and is ready to sign.