AI licence review: identifying IP clauses with confidence

CASUS is a Swiss legal AI platform that analyses licence agreements for IP clauses, prioritises risks by severity, and makes improvement suggestions actionable directly in Microsoft Word. Data is hosted exclusively in Switzerland or the EU, with no transfer to the US. Zero Data Retention and no Human Review are active by default – relevant for confidential technology and pharmaceutical licences.

Why IP clauses in licence agreements are so prone to gaps

Licence agreements are among the most complex contracts in commercial practice. The difficulty is not that individual clauses are hard to understand. It is that many IP-relevant provisions are distributed across the full contract text and only produce their effect when read together. An ownership clause in section 4, an improvement provision in section 12, and a sublicensing restriction in section 18 are logically linked – but easy to miss when reviewed in isolation.

For in-house teams at technology and pharmaceutical companies, this is a real daily challenge. Contracts frequently arrive for review at short notice, time pressure is high, and the variety of clause configurations is wide. Errors in IP clauses are also particularly consequential: incorrect ownership rules for improvements, or poorly drafted back-licences, can result in the loss of strategically important intellectual property rights – with consequences that may materialise in litigation as a damages claim under Art. 97 OR (Swiss Code of Obligations).

An in-house legal team at a Zurich medical technology company using CASUS for out-licensing review reduced the average time to a first complete risk overview of a 30-page licence agreement from roughly four hours to under one hour. The time recovered flows directly into negotiation preparation.

What makes a licence agreement distinctive from an IP perspective?

A licence agreement grants the right to use another party's intellectual property – typically a patent, software, trade mark, know-how, or a combination. Under Swiss law, the licence agreement is not a separately codified contract type; it is interpreted under the general provisions of the OR, in particular Art. 18 OR (interpretation of contracts according to the actual intention of the parties) and Art. 394 ff. OR (mandate law) where a service element is added.

The critical IP clauses address three questions: what exactly is being licensed, under what conditions may it be used, and who owns improvements and further developments?

For software, the starting point under Swiss copyright law (Urheberrechtsgesetz, URG) is essential: Art. 2 para. 3 URG expressly qualifies computer programs as copyright-protected works, provided they meet the general work criterion (individual intellectual creation). Art. 6 URG establishes that only natural persons can be authors – a provision with direct consequences for AI-generated software components (addressed further below). Licensing software IP in Switzerland therefore always operates in the tension between contractual freedom and the mandatory limits of the URG.

Added to this are provisions on sublicensing, confidentiality, termination and its IP consequences (return of materials, deletion obligations, post-termination use rights), warranties for the protected rights, and – for software – source code escrow and deposit mechanisms. Art. 332 OR (rights to inventions and designs created in an employment relationship) is an employment law provision, but it provides a standard of interpretation for the question of who is entitled to development work performed in the course of a licensing relationship. In pharmaceutical out-licensing, this point is routinely underestimated.

In software IP licences specifically, the boundary between the licensed right and the accompanying know-how transfer is often poorly defined, which leads to significant interpretive effort under Art. 18 OR when disputes arise.

How AI changes the review of IP clauses

AI tools now make it possible to review a licence agreement systematically for these clauses without manually calling up each section individually. The practical difference is less about speed and more about completeness: a structured AI review also identifies regulatory gaps – topics that are simply absent from the contract.

CASUS offers two main approaches for this. The Risk & Quality Review analyses the licence agreement from each party's perspective: which clauses are disadvantageous for the client's side, which are incomplete, and which are clearly drafted in favour of the counterparty? Each finding is prioritised by severity (low / medium / high) and paired with concrete drafting suggestions that can be applied directly in Word.

The Benchmark workflow takes a different route: it checks the document against a reference standard – either an internal playbook or established best practices for licence agreements. The result shows which standard clauses are missing, which are present but incomplete, and where deviations from the standard exist. A percentage score reflects overall conformity.

For teams that want to systematically review a licence portfolio, the AI Data Room enables the parallel analysis of dozens or hundreds of licence agreements, with results presented in a structured table – organised by defined fields such as IP ownership, sublicensing restrictions, or post-termination consequences.

Typical IP clause gaps that AI surfaces

In licence agreements analysed through CASUS, a recurring pattern appears: with contracts from SME licensors, the gap in improvement ownership rules shows up in a substantial majority of cases – most frequently in agreements drafted without external legal counsel. In pharmaceutical out-licensing agreements, the most common blind spot is the missing back-licence on termination: the licensee has developed over years on the basis of licensed IP, with no contractual rule on what happens to those developments when the agreement ends.

Concretely recurring gaps:

IP ownership for improvements not defined. The Benchmark workflow flags this explicitly as a missing topic area. Without a clear rule, disputes must be resolved by determining the hypothetical intent of the parties under Art. 18 OR – an expensive exercise.

Sublicensing without restrictions. Without a clear rule on whether and under what conditions further sublicensing is permitted, unintended chains of use can develop. In software licences, this directly raises the question whether the licensed URG right is transferable at all.

No defined procedure for third-party IP infringement. Who must bring an action? Who bears the costs? Who benefits from a successful outcome? In practice, the absence of this provision means that when an infringement actually occurs, the parties must first negotiate cost allocation before any legal steps can be taken.

Back-licences after contract termination not addressed. A licensee who has developed over years on the basis of licensed IP faces an unresolved problem on termination: the licensee's own developments are attributable to them under Art. 6 URG, but are technically inseparable from the licensed IP.

Confidentiality without deletion obligation. Know-how transferred under the licence should be returned or destroyed upon termination. Without this rule, confidentiality remains effectively unsecured – a problem that appears most frequently in licence agreements with an embedded know-how transfer.

Liability exclusion without a cap. Art. 100 OR sets mandatory limits on liability exclusions for gross negligence. Despite this, liability caps are frequently absent from licence agreements, which makes the exclusion vulnerable in a dispute.

Teams looking to try this in practice: a free account is available at app.getcasus.com/signup. The platform is ready to use immediately – through the web app without installation, or as a Word add-in within the usual working environment.

Licence agreement review before and after CASUS: a workflow comparison

The following comparison shows how the review process for a typical 30-page software licence agreement changes in an in-house team – not as a theoretical model, but as a concrete illustration of the step reduction that structured AI support enables.

The time recovered is not primarily an efficiency gain – it is negotiation preparation. Having a complete risk overview earlier allows for a more substantive prioritisation of points to be renegotiated.

On the question of which module to use when: the Risk Review is the right tool when a counterparty contract is in hand and party-aware risk analysis is needed. The Benchmark is appropriate when an internal template or playbook serves as the reference and deviations need to be identified. The AI Data Room comes into play as soon as multiple contracts need to be reviewed in parallel – for example in a due diligence transaction or a portfolio review.

AI-generated content and copyright – a specific IP question

Licence agreements intended to cover AI-generated output raise a distinct legal question: can purely AI-generated content be protected by copyright and therefore licensed at all?

The answer under Swiss URG is clear: no. Art. 2 para. 1 URG defines works as intellectual creations with individual character. Art. 6 URG establishes that only natural persons can be authors. Purely AI-generated output without a sufficient human creative contribution does not meet these requirements and is therefore in the public domain. Classical licence agreements built on the URG simply do not apply to this type of content.

The Swiss Federal Supreme Court has not yet issued a definitive ruling on AI-generated works, but the prevailing academic view is that the author-personality doctrine anchored in Art. 6 URG excludes an AI as author. The Swiss Federal Council stated in its AI strategy of February 2025 that existing law is broadly adequate for AI outputs, but that targeted amendments to the URG are being examined. A consultation draft is being prepared by the end of 2026, and the discussion is progressing – for licence agreement drafters, a signal to contractually secure rights to AI output today, even without a copyright foundation.

Copyright becomes relevant only where a human contributed sufficiently to the creation – for example through substantial post-editing or detailed creative direction that is recognisable in the output. For hybrid works, the principle of separability applies: the human-created elements remain protected under Art. 2 URG, while the purely AI-generated parts are not.

Licence agreements intended to cover AI output therefore require new contractual structures: contractual usage rights without a copyright foundation, combined with know-how protection and confidentiality arrangements referencing Art. 321a OR (employee duty of loyalty, also used as an interpretive benchmark in B2B contexts).

The legal situation for AI contracts themselves: what to watch when purchasing AI

Anyone who purchases AI services enters into a licence or usage agreement of their own – and that agreement has its own IP complications. Many standard contracts from large AI providers are tailored to US law and address Swiss and European requirements only superficially.

The EU AI Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024. General application begins from 2 August 2026, while specific provisions such as prohibited practices (Art. 5) have applied since 2 February 2025. The AI Act does not apply directly to Swiss companies, but it affects every organisation that deploys or distributes AI systems in the EU market. In its AI strategy of February 2025, the Federal Council announced close monitoring of EU legal developments and did not exclude the autonomous adoption of parts of the AI Act – a relevant planning factor for Swiss companies with EU business.

In AI usage contracts, the following clauses deserve particular attention from an IP perspective: who owns the output, may the provider use inputs for training, what IP indemnification exists for third-party claims arising from training data rights, and how is liability for incorrect or rights-infringing AI outputs allocated – and is that allocation compatible with Art. 100 OR?

With the AI Chat, specific questions like these can be put directly to the document. Answers are linked to the relevant passages – enabling targeted navigation without manually searching through the contract.

Data protection in AI-assisted licence review: revDSG and Zero Data Retention

For in-house teams at larger organisations, the data protection dimension is often decisive when choosing an AI tool for contract analysis: anyone uploading confidential licence agreements containing sensitive know-how needs to know where the data resides and whether it is used for model training.

The revised Swiss Federal Act on Data Protection (revDSG / nDSG, in force since 1 September 2023) sets clear requirements. Art. 5 lit. f revDSG defines profiling and thereby the limits of permissible automated processing of personal data. Art. 16 revDSG governs cross-border data transfers: transmission to countries without an adequate level of protection is impermissible without appropriate safeguards (standard contractual clauses, adequacy decision). Since 15 September 2024, the Swiss-U.S. Data Privacy Framework applies – transfers to DPF-certified US companies are permissible without additional safeguards, while transfers to non-certified US recipients still require them. For financial institutions under FINMA supervision, additional requirements apply under FINMA Circular 2023/1 (Operational Risks and Resilience of Banks) and the outsourcing rules, both of which must be assessed for cloud-based AI tools.

CASUS hosts exclusively in Switzerland and the EU, transfers no data to the US, and does not use submitted documents for model training. Zero Data Retention means documents are not stored after the session. No Human Review means no CASUS employee accesses submitted documents. This configuration is compatible with the requirements of Art. 16 revDSG for cross-border data transfers and corresponds to what Swiss pharmaceutical companies and financial institutions need to demonstrate in their contract and outsourcing documentation.

Further information on the security architecture and data residency is available on the security page.

The Proofread module completes the review process at the final stage: it checks cross-references, definitions, and annexes for formal consistency – a step that is particularly error-prone in licence agreements with many internal references, such as those to patent lists as attachments.

FAQ

What can AI concretely do when reviewing a licence agreement?

AI can analyse licence agreements for IP clauses in a structured way, identify gaps and deviations from standard practice, prioritise risks by severity, and produce concrete drafting suggestions. CASUS provides, per finding, an assignment to the affected party, a relevance assessment, and a severity rating (low / medium / high). The legal assessment and negotiation strategy remain the responsibility of the responsible lawyer.

Can purely AI-generated content be protected by copyright?

No. Art. 2 para. 1 URG requires an individual intellectual creation; Art. 6 URG requires a natural person as author. Purely AI-generated output without a sufficient human creative contribution is in the public domain and cannot be licensed under classical copyright law. The Swiss Federal Council indicated in its AI strategy of February 2025 that targeted URG amendments are being examined; a consultation draft is being prepared by the end of 2026.

Which IP clauses are most often missing from licence agreements?

Recurring gaps from CASUS analyses: ownership rules for improvements, sublicensing restrictions, procedures for third-party IP infringement, back-licences after termination, deletion obligations for transferred know-how, and liability caps. A missing liability cap is also vulnerable under Art. 100 OR, which voids exclusions for gross negligence regardless – so the clause protects the party less than it appears to.

When should I use Risk Review, Benchmark, or AI Data Room?

The Risk Review is the right tool when a counterparty contract is in hand and party-aware risk analysis is needed. The Benchmark is appropriate when an internal template or playbook serves as the reference and deviations or gaps need to be identified. The AI Data Room comes into play when multiple contracts need to be reviewed in parallel – for example in a due diligence transaction or a portfolio review.

How does CASUS protect confidential licence agreement data?

CASUS hosts exclusively in Switzerland and the EU, transfers no data to the US, applies Zero Data Retention by default, and does not offer Human Review. Documents are not used for model training. This configuration is compatible with Art. 16 revDSG on cross-border data transfers and with the standard outsourcing requirements of FINMA-regulated institutions.

What is the difference between the Risk Review and the Benchmark for licence agreements?

The Risk Review analyses the contract from each party's perspective and identifies disadvantageous or imbalanced clauses. The Benchmark checks the document against a reference standard and shows which standard clauses are missing or deviate. The two approaches are complementary: a complete licence review typically starts with the Benchmark for completeness, then moves to the Risk Review for party-aware risk analysis.

How does the AI Data Room help with a licence portfolio?

The AI Data Room enables the parallel analysis of many licence agreements. Defined extraction fields – for example IP ownership, sublicensing, post-termination consequences, liability caps – produce a tabular overview that can be used directly for due diligence or compliance purposes. Anomalies, such as missing caps or unusually long notice periods, are flagged explicitly.

Does the EU AI Act apply to Swiss companies?

The EU AI Act (Regulation (EU) 2024/1689) does not apply directly to Swiss companies, but it affects every organisation that deploys or distributes AI systems in the EU market. Prohibited practices under Art. 5 have applied since 2 February 2025. The Swiss Federal Council stated in its AI strategy of February 2025 that it will closely monitor EU legal developments and may consider adopting parts of the AI Act autonomously.

Can CASUS also review English-language licence agreements?

Yes. CASUS works independently of language – English-language licence agreements, as common in international pharmaceutical out-licensing or technology transfer mandates, can be analysed in the same way as German-language documents.

Getting started with AI-assisted licence agreement review

Licence agreements are too specific for a generic contract check to be sufficient. Anyone looking to review IP clauses systematically and with attention to their own negotiating position needs a tool that combines party-aware analysis, completeness checks, and drafting support – and that meets Swiss data protection requirements under revDSG.

CASUS offers the Risk Review for party-aware risk analysis, the Benchmark for comparison against playbooks, and the AI Data Room for portfolio-level reviews. All within an environment hosted exclusively in Switzerland and the EU. Getting started is free at app.getcasus.com/signup. More about the platform and the team on the about page.