ComparisonFrontier R&D
Who owns commissioned research, and who may publish it
Ownership and publication terms decide what you can do with research results long after the work ends, so they belong in the contract before the first experiment. This page compares four arrangements, from fully confidential client ownership to open-source release, on control, patent options, credibility and cost, and explains why any patent filing has to precede a paper or public repository. It is preparation for a conversation with your counsel, not legal advice.
On this page
- Why research terms are fixed before the first experiment
- Background IP, foreground IP and other contract terms
- Four arrangements compared on control, patents and credibility
- Publication versus patents: grace periods differ by jurisdiction
- Choosing a licence and the moment of release
- What a research data-handling schedule should list
- Matching the arrangement to your situation
- Questions and answers
- Sources
Why research terms are fixed before the first experiment
Research produces things of uncertain value: code, datasets, measurements, negative results and occasionally an invention. Once they exist, each party has a view of what they are worth, and allocating them becomes a negotiation with a winner and a loser. Agreed beforehand, the same allocation is a design choice made before anyone knows the outcome, which is why ColdAI writes ownership and publication into engagement terms before work begins1.
Timing matters for a second reason. A paper, a conference talk or a public repository can count as prior art against a later patent application, including one filed by the people who published. Deciding the order of events after the fact may already be too late.
Background IP, foreground IP and other contract terms
- Background IP
- Know-how, code, tools and rights each party held before the engagement or developed outside it. ColdAI keeps its pre-existing background IP whichever arrangement is chosen for the results1.
- Foreground IP
- Everything created during the engagement: experimental code, configurations, trained models, datasets derived from your data, reports and any inventions. The options below allocate this.
- Licence-back
- A right the owner of foreground IP grants the other party, for example to keep using general methods, or to use background tools embedded in the deliverables.
- Review period
- A window in which the client reads a draft paper, removes confidential material and can ask for a delay to file a patent before submission.
- Data-handling schedule
- An annex listing the data provided, where it is stored, who may access it, how long it is kept and how it is deleted or returned.
Four arrangements compared on control, patents and credibility
The first three are the arrangements described on our Frontier R&D page1. Open-source release can stand alone or be combined with the second or third for the code only.
| Criterion | Client-owned, confidential | Client-owned, joint paper after review | Shared ownership | Open-source release |
|---|---|---|---|---|
| Control over results | Complete; nothing leaves without consent | High; you approve the paper and its timing | Divided; each party uses results within agreed limits | Low once released; anyone may use the code under the licence |
| Patent options | Preserved while inventions stay confidential | Preserved if filing happens inside the review period | Possible, but co-owners must agree who files and pays | Largely lost for what is disclosed, unless filed first |
| External credibility | Only your own claims | Peer review or public scrutiny backs the findings | Others can cite and build on the results | Highest; anyone can rerun the experiments |
| Recruiting and reputation | Little, as the work cannot be shown | A named paper signals capability | Visible where the problem is widely shared | Strong among engineers who use the code |
| Effect on cost | Full effort is borne by the sponsor | Both parties gain from the paper | Shared value can justify shared cost | Clean, documented release adds effort |
| Your data | Fully protected | Excluded from the paper unless you approve | Protected by the schedule; results described generally | Must be synthetic, public or left out |
How the cost row plays out is agreed per engagement; it describes where value sits, not a price.
Publication versus patents: grace periods differ by jurisdiction
Whether your own disclosure destroys novelty depends on where you file, and the difference between the two most common routes is large enough to fix the order of events.
35 U.S.C. § 102, as amended by the Leahy-Smith America Invents Act
United StatesApplies whenYou plan a US application covering something disclosed in a paper, talk or repository.
- Under 35 U.S.C. 102(b)(1), a disclosure made one year or less before the effective filing date by an inventor, a joint inventor or someone who obtained the subject matter from them is not prior art against the claimed invention2.
- The exception covers disclosures traceable to the inventors; it is not a general amnesty, so relying on it should be planned with counsel.
European Patent Convention, Article 54 and Article 55
European Patent Organisation member statesApplies whenYou plan to file a European application through the European Patent Office.
- Article 54 treats everything made available to the public before the filing date as prior art, with no general grace period for an applicant's own publications3.
- Article 55 disregards a disclosure only if it occurred no earlier than six months before filing and resulted from an evident abuse against the applicant or from display at an officially recognised international exhibition4.
Choosing a licence and the moment of release
Releasing research code is a publication with legal terms attached. Permissive licences allow reuse with few conditions; copyleft licences require derivative works to be shared on the same terms. Where methods may be patentable, the patent clause matters: the Apache License 2.0 grants an express patent licence from each contributor, which terminates for anyone who sues claiming the work infringes a patent5. Some shorter permissive licences say nothing explicit about patents.
Release timing follows the same logic as papers: after any patent filing, after the client's review, and with data removed or replaced by synthetic or public equivalents. A repository that includes a sponsor's configuration files or logs can disclose more than the paper ever did.
What a research data-handling schedule should list
Matching the arrangement to your situation
- If
Results could be a competitive differentiator or contain an invention you intend to patent.
ThenChoose client-owned and confidential, and plan any filing before results travel further.
Every disclosure narrows patent options, especially in Europe.
- If
You need external credibility with investors, regulators or customers, but the core method is yours.
ThenChoose client ownership with a jointly approved paper after a review period.
The review period protects confidential detail and leaves time to file.
- If
The question is an open problem many organisations share, such as how to benchmark a class of systems.
ThenConsider shared ownership, possibly with the code released.
The value comes from others adopting and extending the results.
- If
You mainly need others to adopt a protocol, format or tool.
ThenPlan an open-source release under a licence chosen for that purpose.
Adoption depends on people using the code without negotiating first.
Questions and answers
Does commissioning research automatically make us the owner of the results?
Not automatically. Ownership follows the contract, and without clear terms the default rules of the governing law decide, which may surprise both parties. Even under client ownership, each side usually keeps its background IP, so the contract should also state what licence you receive to any background tools embedded in the deliverables.
Can we publish a paper and still file a patent?
Often, if the filing comes first. File, then publish. If the paper goes out first, a US application may still be possible within the one-year grace period for the inventors' own disclosures, but a European application generally will not, because the European Patent Convention has no equivalent general grace period. Plan the sequence with patent counsel before submitting anything.
What happens to negative results under a confidential arrangement?
They belong to you like any other result and stay confidential. Some sponsors later agree to publish negative results in general terms, because they save others from repeating the work, but that needs the same written approval as any other publication. Under the joint-paper arrangement, negative results can appear in the paper if the review agrees.
Can ColdAI reuse what it learned on our project?
That depends on the contract, which should state whether general techniques developed during the work may be reused without your data or identifying details. Your foreground IP and confidential information stay protected under whichever arrangement you choose, and ColdAI keeps its pre-existing background IP. Where the line is unclear for a particular project, agree a list of examples during scoping so nobody has to interpret it later.
Sources
- Frontier R&D: ownership, publication and data-handling FAQs — ColdAI
- 35 U.S. Code § 102 – Conditions for patentability; novelty — Legal Information Institute, Cornell Law School · checked 10 October 2026
- European Patent Convention, Article 54: Novelty — European Patent Office · checked 10 October 2026
- European Patent Convention, Article 55: Non-prejudicial disclosures — European Patent Office · checked 10 October 2026
- Apache License, Version 2.0 — The Apache Software Foundation · checked 10 October 2026