IP clauses in commercial contracts

The most valuable asset of many companies cannot be touched. It sits in brands, data, software, processes, and know-how. Economists Jonathan Haskel and Stian Westlake call this the intangible economy: for the first time, the major developed economies invest more in intangibles than in machines and buildings. According to Ocean Tomo’s Intangible Asset Market Value Study, intangibles account for roughly 90 percent of the market value of the S&P 500; in 1975 it was 17 percent.

This kind of asset has a special property that Haskel and Westlake call spillover: its benefits flow easily to others, and proving who owns it is often hard. That is strength and weakness at once. The same process can work in a thousand products at the same time without wearing out; just as easily, it works in other people’s products. A machine can be locked away; a process or a dataset cannot. The fence around intangible assets is not a lock – it is the contract. Allocate the rights to it wrongly, and you lose them, often unnoticed and usually for good.

This rarely happens through theft. It happens through two sentences that surface again and again in negotiations, and both are wrong. “We paid for it, so we own it.” “We are co-owners, so we may use it.” Both confuse two questions a contract must answer separately: who owns a right, and who may do what with it. The wrong answer costs not elegance but market share, bargaining power, and, in the worst case, the indemnity you owe a deep-pocketed customer.

This article shows how a sound IP clause allocates rights between the parties, throughout from both sides, the licensor and the licensee. What artificial intelligence makes of these questions, from training to competing products, I address separately in the article “Data and AI in contracts: what may your counterparty do with your data?”.

Two questions, not one: ownership and right to use

Every IP clause answers two questions, and they are not the same. The first is ownership: who holds title to the right? The second is the right to use: who may use it, and for what? A contract that answers only the first leaves the commercially decisive one open.

Ownership sounds like security, but it carries cost and burden: registration, maintenance, defense, and enforcement. In most situations a party needs no title at all, only permission to do exactly what its business requires. A well-scoped right to use delivers that permission while cost and control stay with a single owner.

Hence the basic pattern of every good IP clause: ownership and use are decoupled. One party becomes sole owner and bears the cost; the other receives a license as far as its business purpose reaches. From the licensor side that means keeping title and drawing the license narrowly. From the licensee side it means not insisting reflexively on ownership, but making sure the right to use is broad and secure enough to carry the investment.

What each party brings: background IP

At the start of every project sits what each side already holds. This pre-existing intellectual property, the background IP, stays in principle with its owner. The other party receives at most a license, and only as far as the project requires. The standard formula: each party keeps its background; the other gets a right to use it within the project, not beyond.

Two points decide the dispute here. First, the boundary: what counts as a modification or improvement of the background, and who owns it? The line between a derivative work and an independent new creation is not clean, so it belongs in the contract. The usual, clean choice is to assign modifications of a party’s background to the party whose background was modified. Second, containment: the right to use someone else’s background must not live on detached from the project results, or it can be carved out of the project and used elsewhere.¹

From the licensor side, the background is to be held firmly and the license confined sharply to the project purpose. From the licensee side, the danger is a project-bound license that ends with the project: anyone who wants to exploit the results afterward needs a background license that covers that later use, not just the creation.

What the project creates: foreground IP

Foreground IP, often called results, is what the project creates. Here the ownership question arises, and the contract decides it, not the payment. Three patterns are common: sole ownership by one party, which bears all the IP costs and licenses the other; full assignment to one party; or joint ownership. Whoever holds title bears the cost.

The most robust pattern is usually the first: sole ownership with one party, combined with a broad, irrevocable license to the other, tailored to its field of business. That gives each side certainty over use without the imponderables of joint ownership. From the licensor side, assigning the foreground to itself plus a license back is attractive; from the licensee side, what matters is that its license back reaches far enough to actually exploit the results it paid to create.

One point regularly slips through: what counts as foreground at all is decided by the definition in the contract. Registered rights such as patents and designs are covered beyond dispute; whether know-how, trade secrets, copyright, and other work results belong there too depends on how the parties frame the term.² That is exactly where the value sits: the larger part of what a development project produces is rarely a registered right and usually knowledge. Take the foreground with a narrow definition and you take the smaller part. Take it with a broad one and you still need the handover: the contract must provide that the know-how is actually conveyed, by handing over the documents, samples, and prototypes that embody it. An allocation clause without a duty to hand over distributes a title nobody can exercise.

The joint IP trap

Joint ownership arises when both sides contribute to the results so that the contributions can no longer be separated. It sounds fair and is the most dangerous allocation, for one reason: only an active creative contribution counts, mere assistance, suggestions, or routine work create no co-ownership, and the statutory default rules diverge from one jurisdiction to the next. Write “jointly owned” and say nothing else, and you import, unintentionally, the default rule of whatever law applies, and that rule is different everywhere.

How far apart shows in the most practical question: may one co-owner license the joint right alone? For copyright, the three legal orders that shape international contracts answer like this:³

The same clause “jointly owned” produces three different legal worlds. A US-minded party that assumes “I am a co-owner, so I may license” infringes the others’ rights in Germany and the United Kingdom.

In technical business a second layer comes on top, and it inverts the German result. Where several persons have jointly made an invention, the right to the patent belongs to them jointly (Section 6 sentence 2 PatG); in contract practice this leads to a community of fractional shares under Sections 741 et seq. BGB.² Its rules run differently from Section 8 UrhG: each co-owner may dispose of its own share alone, while the joint object as a whole can be disposed of only jointly (Section 747 BGB) – yet each may use it so far as it does not impair the joint use of the others (Section 743(2) BGB). Carry the copyright answer over to a patent, and you get it exactly backwards: for a work, solo exploitation is barred; for an invention, one’s own use is permitted, and only disposal of the whole is blocked.

The lesson is the same in both cases: avoid joint ownership where you can, and use sole ownership plus a cross-license instead. Where it cannot be avoided, the contract must expressly govern the six points the defaults treat inconsistently: exploitation rights, consent to licensing, accounting for revenue, filing and maintenance of the registered rights, enforcement and standing, and the resolution of deadlock. The fourth is the one most often forgotten: who files, who bears the cost, and what happens when one party stops paying? Regulate it, and the remaining party may continue the filing in its own name and at its own expense while the departing party keeps its right of use and loses ownership. Leave it out, and you can only watch.¹

Where competition law limits the allocation

Where development is joint or commissioned, the allocation is not a question of bargaining power alone. Agreements on joint research and development and on paid-for research and development fall within the scope of the R&D Block Exemption Regulation (Regulation (EU) 2023/1066), which has applied since 1 July 2023 and runs until 30 June 2035. It exempts such agreements from the cartel prohibition to the extent they contain restrictions of competition within the meaning of Article 101(1) TFEU at all, and it ties the exemption to conditions that hit the IP clause squarely.

The most important one: the agreement must give all parties full access to the final results, for further research and development as well as for exploitation, including the intellectual property rights and know-how arising from them, and as soon as the results are available (Article 3(2) and (3)). Any agreed compensation for that access must not be so high that it practically prevents it (Article 3(4)). Where no joint exploitation is provided for, the same applies to access to the other party’s pre-existing know-how insofar as it is indispensable for exploitation (Article 4). Two common clauses fall outside the exemption: restricting the freedom to carry out further research in the field concerned, independently or with third parties, after the project ends (Article 8(a)), and the obligation not to challenge the validity of the other party’s intellectual property rights (Article 9(1)(a)); a right to terminate in the event of a challenge remains permissible (Article 9(2)). Where the parties are competitors, a market share threshold of 25 percent applies on top (Article 6(1)).⁴

For drafting this means: the pattern “one party becomes sole owner, the other receives a license” remains available, but within the scope of the Regulation the license must not be cut so narrowly that it devalues access to the results. Turn the license back into a formality, or load it with a prohibitive royalty, and you lose not just bargaining capital but the exemption.

Right to use: tailoring the license

Do you even need a right to use?

Before drafting a license, it is worth asking whether you need one. It depends on whether the contract transfers a thing or permits the use of protected intellectual property.

On a pure sale of goods, such as a machine, no right to use is needed. The power to use the machine follows from ownership: the owner may deal with the thing as it pleases (Section 903 sentence 1 BGB). Reselling the device is free too, because the distribution right is exhausted once the item is put on the market (Section 17(2) UrhG; in US law the first-sale doctrine, 17 U.S.C. § 109(a), and patent exhaustion).⁵ It is different for embedded software: buying the machine acquires ownership of the thing, not the copyright in the program. The work stays with the rightholder, and the buyer needs a license (Sections 31, 69c UrhG) to the extent the supplier wants to restrict use beyond the intended use that is permitted anyway (Section 69d(1) UrhG). And with SaaS no copy is handed over at all; the software runs on the provider’s servers.⁶

That is why, from the licensee side, the precise term in SaaS is not a “right to use” a copy but a “right to access and use the service”: there is nothing to reproduce and nothing that could be exhausted or resold. Import the software-license logic here, and you draft past the subject matter.

The levers of a license

Where a right to use is needed, its reach decides its worth. A license is tailored along eight levers: field of use, territory, exclusivity, sublicensing, term, revocability, consideration, and transferability. The principle behind them is hard: whatever is not expressly granted stays reserved.

How sharply these levers cut shows in a typical licensor-side formula that closes every gap: “a limited, non-exclusive, revocable, non-assignable, non-transferable, and non-sublicensable right to access and use”. Every word shifts the risk, and for the licensee every restriction is a breaking point.

One caveat applies to all eight levers: they usually sit in pre-formulated terms, and then they are open to review. Sections 307 et seq. BGB apply to copyright licensing agreements as well, in business dealings essentially through the general clause of Section 307 BGB. What decides the outcome is the yardstick, that is, which statutory model applies: depart from it and the clause is, in case of doubt, invalid under Section 307(2) no. 1 BGB. For copyright licensing agreements the point is settled: Section 31(5) UrhG is not a statutory model but remains a rule of interpretation, and a grant of all rights of use against a lump-sum fee is therefore available, provided the fee is appropriate.⁷ The Federal Court of Justice continued that line and named the test: a provision is a model only where it rests on a fundamental statutory decision in the sense of a requirement of justice.⁸⁹ That relieves the clause without making it safe: as a rule of interpretation, Section 31(5) UrhG still bites wherever the scope is not stated expressly. What carries here is the precision of the stated purpose, not the breadth of the formula. Why the counter-clause, that everything was individually negotiated, does not save it, I set out in the article “No individual agreement”.

Software: object code, source code, and the reverse-engineering ban

With software, a plainer choice comes before the levers: what is handed over at all? Deliver the object code and you deliver the running program; hand over the source code and you hand over the blueprint. From the licensor side, the source code stays in house, in escrow if need be. The usual counterpart is the ban on examining, testing, or reverse-engineering the program.

That ban does not carry as far as it sounds. The CJEU has held that the lawful acquirer of a program may decompile it, in whole or in part, to correct errors that impair its functioning, and that it need not meet the strict conditions of the interoperability exception to do so. The parties may not contractually exclude every possibility of error correction.¹⁰ What they may do is set the modalities, and that is where the practical answer lies: if the supplier takes on corrective maintenance itself, or if the source code is accessible to the acquirer anyway, decompilation is no longer necessary and the ban holds. Whoever only prohibits and offers nothing has the weaker clause. Conversely, the acquirer stays tightly bound: it may not use the result of the decompilation for any purpose other than correcting the errors.

Notes on US copyright: derivative works, work for hire, fair use

Three concepts of US law shape international IP clauses so strongly that they surface even in contracts governed by German law. Knowing how to place them makes for sharper negotiation.

Derivative works: adaptation is an exclusive right

A derivative work is a work based on one or more pre-existing works that recasts, transforms, or adapts them (17 U.S.C. § 101): the translation, the revised database, the newly aggregated map, the integration into a software product. The right to adapt rests exclusively with the author (17 U.S.C. § 106(2)); without a license or a limitation, the adaptation is an infringement.¹¹ Two consequences are regularly overlooked: the adapter’s copyright covers only the material it added itself, not the original taken over (§ 103(b)). And whoever adapts on the basis of an unlawful taking receives no protection at all for the parts taken over (§ 103(a)). A flawed license base thus devalues your own product line twice over.

Work for hire does not apply automatically

In US law the work-for-hire doctrine makes the commissioning party the author, but only within narrow limits: for employees within the scope of their work, or for commissioned works that fall into one of nine statutory categories with a signed agreement.¹² The mere commission does not suffice; the independent contractor is simply not an employee. German law does not know work for hire at all; it works through the grant of rights of use (Sections 31 et seq. UrhG). A cross-border clause must therefore serve both systems and make the choice of law expressly, rather than rely on a figure that exists in only one legal order.

Behind this sits a question that negotiations almost never ask: where does the supplier get the rights it promises? They arise first with the natural person who created the result, and they do not travel on by themselves. For software, Section 69b(1) UrhG closes the gap: where an employee develops a program in the performance of its duties or following the employer’s instructions, the employer holds the economic rights by operation of law, without any act by the employee and without separate remuneration. For contracting relationships the provision does not apply: freelancers and software houses are not covered, and there only the contract carries. With inventions the chain is longer, because the employed inventor – unlike the employed programmer – is entitled to separate remuneration under the German Employee Inventions Act; where several are involved, it must be determined separately for each of them (Section 12(2) ArbnErfG). Where a university is involved, the contract with the university is not enough: the university inventor is a third party whose statutory rights under Section 42 ArbnErfG cannot be contracted away to its detriment, so a declaration from that person is needed.¹³ Anyone promising sole ownership of the foreground should know first whether its own chain holds.

Fair use will not cover it

Whoever adapts without asking hopes, in US law, for fair use (17 U.S.C. § 107): if your own use is “transformative” enough, no license is needed. That hope did not come from nowhere. A study of all reported US transformative-use decisions through the start of 2017 shows how strongly the concept worked: of the decisions in which the court found a transformative use and thereby disposed of the case, 94 percent ended in fair use. And in 2021, in Google v. Oracle, the US Supreme Court held that copying roughly 11,500 lines of declaring code from an application programming interface was fair use, even though Google built a commercial platform of its own on it. What carried the day was the function of the copied lines as an interface, their share of 0.4 percent of the interface as a whole, and the finding that Android was not a market substitute for Java SE.

Two years later the same court narrowed the line in Warhol v. Goldsmith. A new meaning or message alone does not suffice; what matters is the comparison of purposes, and where both works serve the same commercial purpose, the first factor weighs against fair use. The core sentence of the decision: the transformation needed for transformative use must go beyond what already makes a work derivative.¹⁴

Taken together, the two decisions yield a workable rule of thumb but no certainty: work your way into a different market and you have a chance; compete with the original in the same market and you do not. For commercial actors the limitation is therefore no reliable anchor; what is reliable is the lawful provenance of the material and a clean license.

A model clause that bundles the core points on background and foreground. It is deliberately customer-friendly: the receiving party is granted the rights in the foreground; from the provider’s side, reverse the allocation. The highlighted terms are placeholders for the providing and the receiving party, not defined contractual terms; replace them with the contract’s own designations.

Across borders: one right, many legal orders

There is no worldwide copyright. Protection is territorial: each state protects only for its own territory, and what governs is the law of the country for which protection is claimed (the lex loci protectionis, Article 8(1) Rome II Regulation; at treaty level Article 5(2) Berne Convention).¹⁵ That connecting factor cannot be set aside by a choice of law (Article 8(3) Rome II Regulation). A worldwide license therefore engages the IP law of every target country in parallel, and no governing-law clause consolidates that.

In practice the risk concentrates in two places: in the EU and the EEA, where the sui generis database right (Sections 87a et seq. UrhG) protects a substantially invested database independently of copyright, and in the country the data comes from. From this follows a shift in the question, and it hits both roles: the licensee asks “may we use this everywhere?”, while the licensor must ask “can we warrant this worldwide?”. Anyone who promises large licensees worldwide rights and indemnities should limit the warranties to what can truly be warranted: documented provenance per dataset, and otherwise delivery “as is” and a cap on liability. How the associated indemnity is confined from the indemnifying party’s side, I set out in the article “Indemnity: how the supplier can reduce its liability”.

What you should do

1. Separate ownership and the right to use into distinct clauses, and decide deliberately whether you really need title.
2. Assign each right to a bucket: background stays with its owner, foreground is expressly allocated, and you avoid joint ownership. Beyond the allocation, provide for the handover of the know-how, or you are distributing a title with nothing in it.
3. With inventions, check the different default: co-inventors may use the invention alone but may not dispose of the whole alone. The copyright answer does not fit here.
4. Settle your own chain of title before you promise sole ownership: employees, freelancers, and subcontractors are not the same in law.
5. Where development is joint or commissioned, check the access rights under the R&D Block Exemption Regulation before you cut the license back narrowly.
6. Settle first whether you need a right to use at all (a thing, a work, or a service), and choose the word access for SaaS.
7. Tailor the license along the eight levers; whatever is not granted stays reserved. In pre-formulated terms, the precision of the purpose counts, not the breadth of the formula.
8. Do not merely ban reverse engineering; take on corrective maintenance, which is what makes the ban stick.
9. With a US nexus, check the special rules: work for hire requires the signed agreement, and adaptations without a license enjoy no protection of their own.
10. Do not rely on fair use; secure the lawful provenance and the license.
11. Limit worldwide warranties to what you can actually warrant.

Conclusion

A good IP clause does three things: it assigns each right to a bucket, it tailors use precisely, and it limits the warranty to what can be kept. Assign, tailor, secure. Leave any one of the three to chance or to the statutory default, and you give away rights or promise what you cannot keep. These are exactly the clauses I draft and review, for both sides and with an eye to the law that actually governs in a dispute.

Frequently asked questions

Background IP, foreground IP, joint IP: what is the difference?

Background IP is what a party brings into the contract pre-existing; it stays with that party, and the other side gets at most a purpose-bound license. Foreground IP arises only in the project; who owns it is decided by the contract, not by paying. Joint IP is jointly and inseparably created; it is risky, because the statutory default rules diverge from one jurisdiction to the next.

Ownership or right to use: what should I ask for?

Often a well-scoped right to use is enough, because ownership carries cost and burden (registration, maintenance, enforcement). From the licensee side, what counts is that the right is broad and secure enough (term, transferability, revocability); from the licensor side, that it stays narrow. Whatever is not expressly granted stays reserved.

Do I have to grant a right to use when I sell a machine?

For the machine itself, no. Use follows from ownership (Section 903 BGB), and resale is covered by exhaustion (Section 17(2) UrhG). For embedded software, yes, because the copyright does not pass with the thing; that needs a license (Sections 31, 69c, 69d UrhG). With SaaS no copy is transferred at all; there the precise term is access, not the use of a copy.

What is a derivative work?

A work that recasts, transforms, or adapts a pre-existing work (17 U.S.C. § 101), such as a translation, a revised database, or a newly aggregated map. Creating it is the author’s exclusive right (§ 106(2)). The adapter’s protection covers only what it added itself (§ 103(b)); where the taking was unlawful, no protection arises for the parts taken over at all (§ 103(a)).

What happens to licenses in insolvency?

They are not automatically safe. In the licensor’s insolvency, German law lets the administrator elect non-performance of contracts not yet fully performed on both sides (Section 103 InsO); whether a license survives that is not conclusively settled in German law. US law expressly protects the licensee if the contract is rejected (11 U.S.C. § 365(n)). Whoever depends on the right existentially takes precautions: an irrevocable, fully paid-up license, escrow, or indeed title. Insolvency is the one case where ownership is not overrated.

Notes

1. On the boundary of modifications to the background, the containment of the background license, and filing and cost arrangements under joint ownership: European IPR Helpdesk, Fact Sheet “IP joint ownership”, October 2015, pp. 2 et seq., 8. ↩ ↩²

2. On the definition of foreground IP, which depending on the contract also covers know-how, trade secrets, and copyright, and on the community of fractional shares in joint foreground IP: Grohmann/Albrecht, Praxisprobleme im Zusammenhang mit F&E-Verträgen, GRUR-Prax 2021, 588. ↩ ↩²

3. On the divergent default rules of co-ownership in a work: United States, 17 U.S.C. § 201(a) (joint authors as tenants in common; the power to grant a non-exclusive license subject to a duty to account rests on House Report No. 94-1476); Germany, Section 8 UrhG (exploitation only jointly, consent not to be withheld contrary to good faith); United Kingdom, Copyright, Designs and Patents Act 1988, s. 10, s. 173(2). ↩

4. Commission Regulation (EU) 2023/1066 of 1 June 2023 on the application of Article 101(3) of the Treaty on the Functioning of the European Union to certain categories of research and development agreements, OJ L 143, 2.6.2023, p. 9; Articles 2(2), 3, 4, 6(1), 8(a), 9, and 13. ↩

5. 17 U.S.C. § 109(a) (first sale); Impression Products, Inc. v. Lexmark International, Inc., 581 U.S. 360 (2017) (patent exhaustion). ↩

6. Section 17(2) and Section 69c no. 3 UrhG (exhaustion of the distribution right); CJEU, Judgment of 3 July 2012 – C-128/11 (UsedSoft v Oracle), on exhaustion for downloaded program copies; CJEU, Judgment of 19 December 2019 – C-263/18 (Tom Kabinet), no exhaustion for the online making-available of other works. ↩

7. BGH, judgment of 31 May 2012 – I ZR 73/10. ↩

8. BGH, judgment of 17 October 2013 – I ZR 41/12. ↩

9. BGH, judgment of 31 May 2012 – I ZR 73/10 (Honorarbedingungen Freie Journalisten), paras. 13, 15 f.; BGH, judgment of 17 October 2013 – I ZR 41/12 (Rechteeinräumung Synchronsprecher), paras. 9, 13; for context Grohmann, AGB-Recht und IP-Verträge, GRUR-Prax 2019, 27, 29; on the earlier state of the debate Castendyk, Lizenzverträge und AGB-Recht, ZUM 2007, 169, 172 f. ↩

10. CJEU, Judgment of 6 October 2021 – C-13/20 (Top System), paras. 63 to 69 and operative part; on Directive 91/250/EEC, today Directive 2009/24/EC, in German law Sections 69d, 69e UrhG. ↩

11. 17 U.S.C. §§ 101, 103, 106; U.S. Copyright Office, Circular 14: Copyright in Derivative Works and Compilations (2020); in German law Section 23(1) UrhG. ↩

12. 17 U.S.C. §§ 101, 201(b); Community for Creative Non-Violence v. Reid, 490 U.S. 730 (1989). ↩

13. On the reach of Section 69b(1) and (2) UrhG and its non-application to contracting relationships, Hoeren, IT-Recht, lecture script, as of July 2020, pp. 64 f.; on inventor remuneration Section 12(2) ArbnErfG; on the separate declaration of the university inventor under Section 42 ArbnErfG, Grohmann/Albrecht (n. 2), GRUR-Prax 2021, 588, 590 f. ↩

14. Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994); Google LLC v. Oracle America, Inc., 593 U.S. ___ (2021); Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023); on the empirical study Liu, An Empirical Study of Transformative Use in Copyright Law, 22 Stan. Tech. L. Rev. 163 (2019), covering decisions through 1 January 2017. ↩

15. Article 8(1) and (3) Rome II Regulation (Regulation (EC) No 864/2007); Article 5(2) Berne Convention; on the sui generis database right Sections 87a et seq. UrhG (Directive 96/9/EC). ↩

Reference: Poleacov, P. (2026). IP clauses in commercial contracts. INN.LAW. https://inn.law/en/perspectives/ip-clauses/

---