MFN clauses in AI data and content licensing deals

This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.

For the bare definition, see the glossary entry on the most-favored-nation clause. This guide covers how a data partnerships lead scopes, drafts and polices one in a training, fine-tuning or retrieval license. It sits within the AI training data licensing buyer's guide.

Why buyers ask for MFNs in AI data licenses

Buyers ask for an MFN because AI data is priced per deal with no public reference price, so an early licensee carries the risk of overpaying relative to later ones. Content and dataset owners often license the same corpus to several model developers over a few years, and the second or third deal frequently lands on different economics once the licensor learns what the market will bear.

An MFN is a hedge on that information gap. It is most valuable when you are an early buyer, when the licensor plans a non-exclusive program, and when your deal has a long term or a renewal tied to the original price. It is least valuable when the asset is a one-off custom collection that no one else will license on the same scope.

The trade-off is leverage. A licensor that grants an MFN constrains every later negotiation, so expect to pay for it with a narrower comparison set, a cap on how far back the adjustment reaches, or a higher base price. Compare that cost against the structures in AI data license pricing structures compared before deciding it is worth the concession.

Defining "comparable terms" for a dataset license

The comparison set decides whether the MFN ever triggers, so define it by the attributes that actually move the price of training data. A licensor will argue that every other deal is different; a vague "similarly situated licensee" standard hands them that argument.

The attributes that matter in AI data deals are:

• Record scope. The same collection, a defined subset, or a superset. Specify whether a deal for 80 percent of your records counts.
• Volume and refresh. One-time snapshot versus ongoing feed, and record or token counts. Per-record and per-token deals need a normalization rule.
• Use rights. Pre-training, fine-tuning only, evaluation, or retrieval-augmented generation. A fine-tuning-only license is not a fair comparator for a pre-training rights grant.
• Exclusivity. Exclusive, field-exclusive or time-limited exclusive deals usually carry a premium and are typically carved out.
• Term, territory and model rights. Length, renewal, and whether derivative and successor model rights survive termination.
• Licensee class. Academic and nonprofit grants, affiliates of the licensor, and settlement licenses are commonly excluded.

Write the normalization rule into the clause. For example, compare effective price per licensed record for the same use category and term band, adjusted pro rata for months of term.

Price-only MFN versus whole-package MFN

A price-only MFN compares one number, while a whole-package MFN compares the full bundle of rights, warranties and remedies, and each fails in a different way. Choose based on which risk you are actually trying to manage.

Illustrative example: invented to show structure; it does not describe an available dataset.

Most buyers land on elect-the-package for price and a separate term-specific MFN for one or two protective clauses, such as the IP indemnity cap. That combination is easier to verify than a whole-package standard and harder to evade than price-only.

Notice, certification and audit mechanics

Verification is the part most MFN clauses get wrong, because the buyer cannot see the licensor's other contracts. The practical answer is a layered mechanism: event-driven notice, periodic officer certification, and a narrow audit right exercised by an independent party.

Illustrative example: invented to show structure; it does not describe an available dataset.

MFN COMPLIANCE MECHANICS (sample structure for counsel to adapt)
1. Trigger:     Licensor grants a Comparable License on More Favorable Terms
                after the Effective Date.
2. Notice:      Written notice within <30> days of execution, stating the
                adjusted terms offered to Licensee (not the other licensee's
                identity or contract).
3. Election:    Licensee has <60> days to elect the adjusted terms in full.
4. Adjustment:  Applies from the date of the other license; credit against
                next invoice, not cash refund.
5. Certificate: Annual certificate signed by an officer of Licensor stating
                either "no trigger occurred" or listing adjusted terms.
6. Audit:       Once per 12 months, by an independent accountant under NDA,
                who reports only "compliant" or the adjustment amount.
7. Lookback:    Limited to licenses executed in the prior <24> months.
8. Exclusions:  Exclusive grants, academic/nonprofit grants, affiliates,
                litigation settlements, pilot evaluations under <X> value.

The independent accountant in step 6 matters for two reasons. It lets the licensor honor confidentiality obligations to its other licensees, a tension covered in naming your data sources: confidentiality versus transparency, and it keeps competitively sensitive pricing out of your hands. Ask that the auditor report a conclusion and an adjustment figure, never the underlying contract.

Common failure modes to plan for:

• The licensor restructures a later deal as a revenue share so there is no comparable flat price; address this with a normalization rule, and see revenue-share data deals.
• The later deal is signed by an affiliate or a newly formed entity; extend the MFN to licensor affiliates and successors.
• The licensor bundles the data with unrelated services; require allocation of bundled consideration in the certificate.

Antitrust limits on MFNs and information exchange

MFNs are generally lawful, but they draw antitrust scrutiny when a party with market power uses them to raise rivals' costs or to coordinate prices, and verification must not turn into competitors sharing pricing. US antitrust analysis usually treats MFNs under the rule of reason, so the market position of the parties and the clause's effect on rivals carry the weight.

Two well-known US enforcement matters show the risk patterns. In United States v. Blue Cross Blue Shield of Michigan (2010), the government challenged "MFN-plus" clauses that required hospitals to charge rival insurers more than the dominant insurer. In United States v. Apple (2013), MFN provisions in e-book agency agreements were part of conduct the court found to be an unlawful pricing conspiracy among publishers and Apple.

For AI data buyers, the lessons translate directly:

• Avoid "MFN-plus" language that requires the licensor to charge other AI developers more than you.
• Do not use an MFN to receive other licensees' prices, identities or terms; route verification through an independent auditor.
• Be more cautious if you are a dominant buyer of a data category or if several model developers seek parallel MFNs from the same licensor.

The guidance landscape has also thinned. DOJ withdrew three health care antitrust policy statements, including the information-exchange safety zone, on February 3, 2023 [2], and DOJ and the FTC later withdrew the Antitrust Guidelines for Collaborations Among Competitors [1]. As of October 2026, that leaves fewer bright-line safe harbors, which is a reason to involve antitrust counsel before signing MFNs in concentrated data categories.

Negotiating positions and fallbacks

Open with the clause you want, but know which concessions keep most of its value. The sequence below fits most non-exclusive training and retrieval licenses.

Illustrative example: invented to show structure; it does not describe an available dataset.

Fold the agreed position into your data license term sheet and the broader negotiation checklist so the MFN is tested against termination, renewal and deletion and return clauses together. If the deal covers retrieval content, check the MFN against the use-based pricing in RAG content license terms.

How MFNs interact with sourcing on request

An MFN assumes the licensor runs a repeatable licensing program with several comparable licensees, which is not always the shape of operational data deals. When data is sourced on request from a company's internal systems, the scope, preparation and permitted uses are often specific to one buyer, so a comparison set may never form.

SourceX sources operational datasets from US companies on request rather than holding them in stock, and every release is approved by the supplying company. Pricing and allowed uses are agreed per deal in a license, and SourceX does not publish prices. Buyers with a specific need can start by describing the data they need to SourceX, and raise pricing protections with their counsel when terms are negotiated.

License operational data with negotiated terms

SourceX sources operational datasets from US companies on request and manages the commercial process, from assessing data and licensing permissions to agreeing pricing and allowed uses in a license. Nothing is contracted until a supplier agrees, and terms are agreed per deal. Describe the data you want to license.