Will an AI vendor train on the data we submit?

Under many default terms, yes — submitted inputs can be retained and used to improve the provider's models. Enterprise tiers often let you turn that off, but only if the contract says so. We negotiate zero-retention and zero-training terms so your proprietary inputs stay out of the vendor's training set.

Who owns the output our team generates with the model?

It depends on the agreement. We structure terms so the enterprise retains ownership of its generated outputs, even where the underlying model weights remain the vendor's — and so output you rely on commercially is not subject to the vendor's reuse.

What is an IP indemnity, and why does it matter for AI?

An indemnity shifts the cost of a third-party infringement claim to the party that agreed to bear it. Because some models were trained on copyrighted material, output can draw infringement claims. We negotiate the vendor's IP indemnity — its scope and any caps — so that risk does not default to you.

Who is liable if the model produces something false or harmful?

Vendors routinely disclaim liability for model output. We allocate that risk in the contract through representations, liability terms, and use restrictions, and pair it with internal review checkpoints so an error is caught before it causes harm.

Do we need anything beyond the contract?

Yes — an internal use policy that aligns how employees actually use the tool with what the contract permits. A strong agreement is undermined if staff paste confidential data into a consumer version. We align the policy with the negotiated terms.

Practice - AI Law - Model procurement

AI Model Procurement & Licensing

When an enterprise integrates a third-party AI model, the default terms often grant the vendor quiet access to proprietary data. We negotiate enterprise procurement agreements that explicitly quarantine your data from public training sets, allocate liability for algorithmic hallucinations, and secure thorough IP indemnification. We treat every API integration as an intellectual property transaction.

Begin the matter Review the steps

Matter: Vendor licensing
Register: Transactional
Counsel: Christopher Moye

Model procurement

Fencing the data

If you do not explicitly protect your inputs in the vendor contract, you have surrendered them.

The problem

An enterprise AI contract signed on the vendor's default terms often gives away your data and leaves you holding the infringement risk.

Standard model agreements frequently let the provider retain and train on submitted inputs, disclaim liability for what the model outputs, and cap or exclude IP indemnities. Each of those is negotiable — but only before signing, which is why the procurement review is where the protection is won.

Principles Â· 01

How we draft the matter.

Every engagement is composed against these commitments. They shape the protections we add, the questions we ask, and the document that leaves the file.

Â§ 01

Fencing the data

If you do not explicitly protect your inputs in the vendor contract, you have surrendered them. We mandate absolute data quarantine.

Â§ 02

Allocating hallucination risk

When the model errs, liability must be apportioned. We draft the indemnities that protect the enterprise from third-party output claims.

Â§ 03

SLA enforcement

Compute is a utility. We negotiate firm uptime commitments and compute-availability clauses so the model performs at scale.

What we watch Â· 02

What can break the matter.

These are the terms, structures, and practical risks that usually decide whether the work holds when the file is tested.

CTODATA OFFICER

Training Data Quarantine

Drafting strict zero-retention and zero-training covenants to ensure enterprise inputs are never used to improve the vendor's foundational models.

FOUNDERGC

Output Ownership

Structuring terms to ensure the enterprise retains exclusive ownership of all generated outputs, even when underlying model weights are proprietary.

ENTERPRISEBOARD

Infringement Indemnification

Negotiating strong, uncapped indemnities from the vendor against claims that the model's outputs or training data infringe third-party IP.

The work Â· 03

Four steps. One engagement.

Each step is concrete; each step has a deliverable. The scope is defined, the matter moves, and the file closes.

01
Term Analysis
We dissect the vendor's standard enterprise agreement to expose data retention rights and liability disclaimers.
02
Redlining & Negotiation
We negotiate for custom data-quarantine provisions, output-ownership clauses, and strict SLA requirements.
03
Indemnity Structuring
We mandate IP infringement indemnification, shifting the risk of training-data copyright violations back to the model provider.
04
Deployment Governance
We establish internal usage policies that align employee interaction with the finalized vendor terms.

Proof

What stands behind the work.

What stands behind the work — credentials and representative engagements, stated plainly.

Authorship

AI procurement matters are handled by Christopher Moyé, Esq., who authors the firm's published writing on AI vendor contracting.

Scope of practice

Enterprise AI vendor agreements — data-quarantine and zero-training terms, output ownership, IP indemnification, and service levels.

How the work is run

We treat every model integration as an IP transaction, reviewing what the contract lets the vendor do with your data before you sign.

Common questions

Questions clients ask.

Plain answers to the questions that come up most. If yours is not here, send the facts — we answer in writing.

Will an AI vendor train on the data we submit?: Under many default terms, yes — submitted inputs can be retained and used to improve the provider's models. Enterprise tiers often let you turn that off, but only if the contract says so. We negotiate zero-retention and zero-training terms so your proprietary inputs stay out of the vendor's training set.
Who owns the output our team generates with the model?: It depends on the agreement. We structure terms so the enterprise retains ownership of its generated outputs, even where the underlying model weights remain the vendor's — and so output you rely on commercially is not subject to the vendor's reuse.
What is an IP indemnity, and why does it matter for AI?: An indemnity shifts the cost of a third-party infringement claim to the party that agreed to bear it. Because some models were trained on copyrighted material, output can draw infringement claims. We negotiate the vendor's IP indemnity — its scope and any caps — so that risk does not default to you.
Who is liable if the model produces something false or harmful?: Vendors routinely disclaim liability for model output. We allocate that risk in the contract through representations, liability terms, and use restrictions, and pair it with internal review checkpoints so an error is caught before it causes harm.
Do we need anything beyond the contract?: Yes — an internal use policy that aligns how employees actually use the tool with what the contract permits. A strong agreement is undermined if staff paste confidential data into a consumer version. We align the policy with the negotiated terms.