When the contract is silent on AI output: what to do first
When a technology contract says nothing about AI output, treat the silence as a drafting gap, not a legal vacuum. Identify the one right that is missing — ownership, licence to use, or liability for errors — and close it before the output reaches a customer-facing product. Waiting turns a fixable gap into a dispute for the contract's named forum to resolve.
Which right is actually missing
"AI output" covers several different things, and a contract usually goes silent on all of them at once: who owns the output, what licence the client actually has to use it, and who carries liability if the output is wrong. The three gaps do not close with the same fix, so treat them as three separate questions rather than one broad concern. A related failure mode — where a party loses practical control over a contract's own terms once a system starts acting on them — is covered in When the issuer loses control of the contract. The mechanics differ, but the discipline of separating the gaps is the same.
| Type of AI output | What is typically missing | Default position without an amendment |
|---|---|---|
| Generated text, reports, summaries | Ownership assignment, warranty of originality | Ownership follows whichever party the underlying agreement already treats as author — rarely stated for AI output specifically |
| Generated source code | Rights in the resulting computer program, scope of licence | Assessed as a computer program under Directive 2009/24/EC, following the same authorship logic as human-written code |
| Generated designs, layouts, images | Licence to use, clearance of any third-party material behind training or prompts | No licence exists by default; use without one is a contractual and IP risk, not merely a drafting oversight |
| Structured recommendations or decisions | Liability cap, audit trail for how the output was produced | Falls back to the general liability clause of the contract, not to any AI-specific standard |
Why this matters under EU rules
Where the AI output is source code, the relevant EU basis is Directive 2009/24/EC. It sets the baseline for how a computer program is protected, regardless of whether a human or a system produced the first draft. The directive does not allocate ownership between client and supplier — the contract still does that — but it fixes what kind of right is at stake once ownership is settled.
Separately, Regulation (EU) 2024/1689 (AI Act) enters into force on 2 August 2026. It governs obligations on providers and deployers of AI systems — risk classification, documentation, transparency — not the ownership of what the system produces. A contract that is silent on output ownership is not fixed once the AI Act becomes applicable. The two questions run on separate tracks, and the second does not answer the first.
What changes in risk and timing
Until the gap is closed, the practical risk sits with whichever party first relies on the output commercially, not with whoever drafted the contract. Timing matters more than fault: an amendment signed before deployment is negotiation; the same amendment requested after a customer complaint is leverage the other side did not have before.
What to do first
- Legal or contract owner: map which of the three rights is actually silent — ownership, licence, or liability — before drafting anything.
- Legal: check whether the output is a computer program within the scope of Directive 2009/24/EC; if it is, apply the same ownership logic your organisation already uses for code written by staff or contractors.
- Procurement or product owner: pause commercial reliance on the output until the gap is closed by an amendment or side letter.
- Legal: draft a short amendment covering three points only — ownership, licence scope, liability cap — rather than reopening the whole contract.
- Authorised signatories on both sides: execute the amendment before the output is used in a customer-facing product or decision.
- Compliance: keep the amendment on file with a dated note. Once the AI Act applies from 2 August 2026, documentation of how output rights were settled becomes part of the compliance record for in-scope systems.
Who decides
There is no separate forum for "AI output disputes." The body that decides is whichever court or arbitral tribunal the underlying contract already names in its governing-law and dispute-resolution clause. Where the output is source code, that forum assesses ownership under the law it applies to computer programs generally, informed by Directive 2009/24/EC where that law implements it — not by any AI-specific standard.
If you are in one of these situations
- If the output is source code already integrated into a live product, treat it under the same rules your organisation applies to any computer program under Directive 2009/24/EC, and confirm assignment before the next release, not after.
- If the AI system that produced the output will fall within the scope of the AI Act once it applies from 2 August 2026, keep provenance records now — retrofitting them later is harder than logging them at the point of generation.
- If the contract has no governing-law clause at all, the ownership question is decided by conflict-of-laws rules before it is decided by either party's preferred jurisdiction. Escalate to counsel rather than assume.
What this page does not solve
This page identifies the missing right, the first steps, and the deciding forum. It does not draft the amendment, and it does not establish, in a live dispute, who a counterparty actually is or what a given document set can prove.
For how counterparty status is established and where that stops being reliable, see Counterparty in Finland and its limits explained.
For how a document itself is assessed once the dispute reaches evidence, see Document in Finland and its limits in practice.
Verified
Checked against the current text of Directive 2009/24/EC and Regulation (EU) 2024/1689 (AI Act) as of 5 June 2025. Rules in this area change; recheck the applicable date before relying on the AI Act reference. For how consistently a same-day answer is delivered on situations of this kind, see Delivery statistics, published automatically in practice.
Related situations
A neighbouring drafting gap, in a different part of the same practice, is addressed in When a migration is blocked by an exit clause.
FAQ
What is the first step when the contract is silent on AI output?
Identify which single right is missing — ownership, licence to use, or liability for errors — before drafting any fix. The three gaps need different clauses, and treating them as one problem produces an amendment that closes the wrong gap.
Which deadline is already running?
Two clocks matter, and neither is the date the contract was signed. The general limitation period for a contractual claim, set by the law governing the contract, starts running from the point the silence first caused loss. Separately, if the AI system in question is in scope of the AI Act, the transition toward its application from 2 August 2026 is already running regardless of what the contract says.
Who decides, and on what evidence?
The forum named in the contract's own dispute-resolution clause decides, on the evidence of the contract itself, any amendment made, and the record of how the output was produced. Where no such record exists, the party that kept none carries the practical disadvantage, whatever the formal burden of proof.