Who Owns Ai‑generated Inventions and Outputs in Denmark? Patents, Copyright, Trade Secrets & Contracts (2026 Guide)

Last reviewed: July 6, 2026

The question of who owns AI-generated works in Denmark has moved from academic debate to boardroom urgency. Denmark’s 2025–2026 legislative proposals on deepfake protections and face-and-voice rights, combined with the European Parliament’s ongoing evaluation of copyright rules for artificial intelligence, mean that every company developing, deploying or licensing AI outputs now faces concrete ownership and compliance decisions. Danish law does not grant copyright to a machine, it does not permit an AI system to be named as a patent inventor, and it offers no single statutory framework that neatly governs the full lifecycle of an AI model’s training data, weights and outputs.

For businesses operating in Denmark, the practical consequence is that ownership of AI output depends almost entirely on the interplay between existing IP statutes, regulatory guidance from the Danish Data Protection Agency (Datatilsynet), and, critically, the contracts a company puts in place before development begins.

Executive Summary, Key Facts for Decision‑Makers

• Copyright. Under the Danish Copyright Act (Ophavsretsloven, consolidated act no. 1093 of 2023), copyright protection requires a human author who has made original creative choices. Purely AI-generated outputs with no qualifying human creative input fall outside copyright protection.
• Patents. Danish patent law and European Patent Office (EPO) practice require that an inventor be a natural person. AI cannot be named as an inventor, but AI-assisted inventions remain patentable provided a human inventor’s contribution meets the inventive-step threshold.
• Deepfake proposals (2025–2026). Denmark has proposed national amendments addressing face-and-voice deepfake protections, a development the European Parliament’s EPRS analysed in January 2026 (EPRS 782611). These proposals may create new consent and labelling obligations for commercial use of synthetic images and voices.
• Contracts are the primary safeguard. Because statutory IP protection for AI-generated works in Denmark is limited and uncertain, well-drafted assignment, licensing and indemnity clauses remain the single most reliable tool for securing commercial rights.

Quick Legal Map, How Danish and EU Law Treat AI Outputs

Understanding ownership of AI output requires mapping each available IP regime to the specific asset in question, the training data, the model architecture, the trained weights, or the final output. No single Danish or EU instrument covers all four. The table below provides a working framework for in-house counsel and IP managers evaluating how to protect AI models in Denmark.

Relevant Danish Statutes and EU Instruments

• Ophavsretsloven (Danish Copyright Act, consolidated act no. 1093 of 2023), governs authorship, economic rights and moral rights.
• Danish Patent Act (Patentloven), implements the European Patent Convention (EPC) requirements at national level; administered by the DKPTO.
• Danish Trade Secrets Act, implements EU Directive 2016/943 on trade secrets.
• Directive (EU) 2019/790 (CDSM Directive), Articles 3 and 4 establish text-and-data mining (TDM) exceptions relevant to training data rights.
• GDPR and Datatilsynet AI guidance, governs processing of personal data in training datasets and model outputs.
• WIPO AI & IP Policy Conversations, international guidance informing Denmark’s and the EU’s regulatory approach to AI-generated works.

Patentability and AI Inventorship in Denmark

The short answer on AI inventorship in Denmark is clear: an AI system cannot be named as an inventor. Under both the Danish Patent Act and EPO practice, an inventor must be a natural person. However, this does not mean AI-assisted inventions are unpatentable, it means the human being who directed, designed or adapted the AI tool to produce the inventive result must be identified and documented as the inventor.

Inventorship, Can AI Be Named?

The DKPTO follows EPC requirements, which mandate designation of a natural person as inventor. This position aligns with the EPO’s decisions refusing patent applications that listed an AI system (DABUS) as the sole inventor. The critical question for businesses is not whether AI “created” an invention but whether a human contributor made an inventive contribution that satisfies the inventive-step test. If the human’s role was limited to pressing a button and reviewing output, inventorship may be difficult to establish. If the human defined the problem, selected and configured the AI tool, chose training data, iterated on parameters, and identified the inventive solution from among the AI’s outputs, that human is the inventor.

Businesses developing AI-assisted inventions should implement contemporaneous documentation practices: lab notebooks, version-control logs, decision memos and meeting minutes that record each human decision in the inventive process. This documentation is essential both for patent prosecution and for defending inventorship in any subsequent challenge.

Practical Prosecution Checklist

• Prior art search. Conduct thorough searches before filing, AI tools can generate solutions that exist in prior art.
• Inventorship declaration. Identify every natural person who made an inventive contribution; obtain signed declarations.
• Chain-of-inventorship records. Maintain logs showing which human decisions shaped the AI’s configuration and output selection.
• Employee inventor clauses. Under Danish law, employee inventions may vest in the employer if the invention falls within the employee’s duties, confirm this is addressed in employment agreements.
• Assignment agreements. Where contractors or third-party AI vendors are involved, ensure written assignments transfer patent rights to the commissioning entity.

Copyright and Denmark’s 2025–2026 Deepfake and AI Copyright Proposals

Copyright protection for AI-generated works in Denmark hinges on whether a human author’s free creative choices are reflected in the work. Under the Ophavsretsloven, the creator (ophavsmand) must be a physical person. Works produced entirely by an AI system, without qualifying human creative input in the expression itself, fall outside the scope of copyright protection. This is consistent with broader Nordic and EU copyright doctrine, which has historically required human authorship as a threshold condition.

The practical implication is significant: if a company uses a generative AI tool to produce marketing copy, design assets or music, and no human author has made original creative choices in shaping the specific expression of the output, that output is likely not protected by copyright AI Denmark rules. It may be freely copied by competitors unless other protections (trade secret, contract, unfair-competition rules) apply.

Where a human author uses an AI tool as an instrument, making creative decisions about prompts, selecting from outputs, editing, combining and refining the result, the resulting work may qualify for copyright protection. The key test is the degree and nature of human creative involvement in the final expression, not merely in initiating the process.

Denmark’s Deepfake Proposals and the EU Policy Arc

Denmark’s 2025–2026 legislative proposals on face-and-voice deepfake protections represent one of the most commercially significant developments for AI-generated works in Denmark. The European Parliament’s EPRS analysis (EPRS 782611, January 2026) examined Denmark’s approach as a notable national initiative within the broader EU copyright and AI policy landscape. Industry observers expect these proposals to create new consent-based obligations for the commercial use of synthetic likenesses, potentially requiring platforms and advertisers to obtain explicit permission before using AI-generated reproductions of a person’s face or voice.

For detailed analysis of Denmark’s deepfake legislation, see our coverage of the Denmark deepfake law.

Business Scenarios, Commercial Implications

• AI-generated marketing visuals. If no human creative choices shaped the specific visual expression, copyright may not attach. Secure rights by contract with the AI platform vendor and maintain trade-secret protections over any proprietary prompts or configurations used.
• Synthetic voice cloned from an employee. Under Denmark’s emerging proposals, the employee’s voice may attract new personality-based protections. Obtain written consent and a clear licence from the individual before commercial deployment. Update employment contracts to address AI voice and likeness rights.
• Automated image generation for e-commerce. Where outputs are generated at scale with minimal human intervention per image, copyright protection is unlikely for individual outputs. Protect the generation pipeline (prompts, configurations, fine-tuned models) as trade secrets and contractually restrict downstream use.

Trade Secrets, Training Data and Protecting AI Models in Denmark

For many businesses, trade-secret protection is the most practical and immediate way to protect AI models in Denmark. The Danish Trade Secrets Act (implementing EU Directive 2016/943) protects information that is secret, has commercial value because it is secret, and has been subject to reasonable steps to maintain secrecy. Model weights, proprietary training pipelines, hyperparameter configurations and internal evaluation benchmarks can all qualify as trade secret AI models, provided the company maintains rigorous confidentiality measures.

Practical steps to establish and maintain trade-secret protection include:

• Access controls. Restrict access to model weights, training code and datasets to authorised personnel on a need-to-know basis.
• Technical measures. Encryption at rest and in transit; secure compute environments; logging of access events.
• Contractual measures. Non-disclosure agreements with every employee, contractor, vendor and research collaborator who accesses proprietary AI assets.
• Internal policies. Classification and labelling systems for confidential AI assets; onboarding and exit procedures for personnel.

Datatilsynet Precedents and Privacy Obligations

The interaction between data protection and AI model development is an area of active regulatory attention in Denmark. The Datatilsynet has issued guidance on AI and has addressed the publication of datasets and AI models containing personal data. Businesses must ensure that training data rights include a lawful basis under the GDPR for processing any personal data, that data subjects’ rights (including erasure) can be operationalised, and that publication or sharing of models does not indirectly expose personal information embedded in the model’s parameters.

The Datatilsynet’s regulatory posture makes it essential for companies to conduct data-protection impact assessments (DPIAs) before training models on datasets that include or may include personal data, and to document the legal basis for processing at each stage of the data pipeline.

Contractual Clauses for Training Data Acquisition

• Warranty of rights. The data provider warrants it holds all necessary rights (copyright, database rights, data-subject consents) to licence the data for AI training purposes.
• Indemnity. The data provider indemnifies the licensee against claims arising from defects in title or third-party rights in the training data.
• Audit rights. The licensee retains the right to audit the provenance and compliance status of supplied training data.
• Opt-out compliance. The agreement addresses the data provider’s obligations regarding CDSM Directive Article 4 opt-out mechanisms (relevant for training data rights sourced from published works).

Contracts, Licensing and Practical Steps, What In‑House Counsel Must Do Today

Because statutory IP protection for AI-generated works in Denmark is limited, contracts are the primary mechanism for establishing, transferring and monetising rights. AI licensing in Denmark requires careful drafting across multiple agreement types: employment contracts, contractor agreements, vendor terms, platform licences and customer-facing output licences.

The following clause categories should appear in every AI-related agreement:

• Output ownership assignment. “All Outputs generated using the Licensee’s Proprietary Model, including derivative data, shall vest in the Licensee upon creation.”
• Training data representations. “The Provider represents and warrants that all Training Data has been lawfully obtained and may be used for machine-learning purposes without infringing third-party rights.”
• Moral rights acknowledgement. Where a human contributor’s creative input may attract moral rights under the Ophavsretsloven, include a waiver (to the extent permitted by Danish law) or an acknowledgement that moral rights remain with the creator but will not be exercised to restrict commercial exploitation.
• Confidentiality and trade-secret preservation. Detailed obligations that align with the reasonable-steps requirement of trade-secret law, specify what is confidential, how it must be stored, and when it must be returned or destroyed.
• Source-code and model escrow. For critical AI assets, escrow arrangements ensure business continuity if a vendor becomes insolvent or ceases support.

Licensing Models, Exclusive, Non‑Exclusive and Platform‑to‑Customer

The licensing structure depends on the commercial model. Exclusive licences grant a single licensee the right to use and exploit the AI output, often justified where the output is custom-developed. Non-exclusive licences are standard for platform-to-customer relationships where the same model serves multiple users. In platform agreements, clearly define whether the customer owns outputs generated using the platform’s model, or whether the platform retains a licence to use those outputs (e.g., for model improvement). Ambiguity here is a frequent source of commercial disputes.

M&A and IP Hygiene, Due Diligence Checklist for AI Assets

Acquirers evaluating targets with significant AI assets should verify:

• Chain of title. Written assignments or employment-agreement provisions covering all contributors (employees, contractors, data annotators).
• Training data provenance. Documentation of data sources, licences, opt-out compliance and data-subject consent records.
• Trade-secret measures. Evidence that the target has maintained reasonable confidentiality steps, access logs, NDA records, classification systems.
• Patent portfolio. If AI-assisted inventions are patented, verify that inventorship designations are defensible and that assignment chains are unbroken.
• Regulatory compliance. Datatilsynet correspondence, DPIAs and records of any supervisory authority engagement regarding AI or data-processing activities.

Enforcement, Remedies, Risk Management and Compliance

Enforcing rights in AI-generated works in Denmark requires a multi-regime approach. Copyright infringement claims are available only where the work qualifies for protection, i.e., where human authorship is established. Patent infringement follows standard Danish and EPO enforcement pathways. Trade-secret misappropriation claims under the Danish Trade Secrets Act allow for injunctive relief and damages, often the most effective tool where proprietary model information has been leaked or reverse-engineered.

Cross-border enforcement is a growing concern. AI models trained in one jurisdiction, deployed in another and generating outputs consumed globally create complex jurisdictional questions. Within the EU, Directive 2004/48 (the Enforcement Directive) provides a baseline for IP enforcement harmonisation, but practical coordination across Member States remains challenging. For a broader discussion of cross-border IP strategy, see our guide on how to protect your intellectual property across borders.

When to Litigate vs. Negotiate

• Litigate when the infringement is clear, damages are quantifiable, and a court order (injunction) is needed to prevent ongoing harm, particularly for trade-secret misappropriation where speed is critical.
• Negotiate when the parties have an ongoing commercial relationship, the rights position is ambiguous (common with AI-generated outputs), or a licensing arrangement would be more valuable than adversarial proceedings.
• Escalate to regulatory complaint when the dispute involves personal-data misuse in training data or model outputs, the Datatilsynet can impose administrative measures and fines under the GDPR.

Internal risk management should include regular IP audits of AI assets, insurance reviews (confirm whether existing professional-liability and cyber policies cover AI-related IP claims), and board-level reporting on AI ownership and compliance posture.

Legislative and Policy Timeline, Key Milestones for AI‑Generated Works in Denmark

For additional context on Denmark’s specific deepfake legislative developments, see our detailed analysis of the Denmark deepfake law. For broader international IP strategy, consult the International Intellectual Property practice guide.

Conclusion, Practical Checklist for AI‑Generated Works in Denmark

The ownership landscape for AI-generated works in Denmark is defined more by what the law does not protect than by what it does. Copyright does not cover AI-only outputs; patent law does not accept AI as an inventor; and statutory trade-secret protection requires active, ongoing measures. Contracts fill the gap, and fill it only if they are in place before development begins.

• Next 30 days: Audit all existing AI-related contracts for output ownership, training data warranties and confidentiality clauses. Identify and remediate gaps.
• Next 90 days: Implement contemporaneous inventorship documentation procedures for AI-assisted R&D. Conduct DPIAs for all training datasets containing personal data.
• Next 180 days: Monitor Denmark’s deepfake legislative progress and the EU Parliament’s CDSM evaluation. Update consent workflows for any commercial use of synthetic likenesses. Brief the board on the company’s AI IP posture and residual risk exposure.

Sources

1. European Parliament, EPRS briefing: The Danish approach to copyright and deepfakes (EPRS 782611, January 2026)
2. EUR-Lex, Directive (EU) 2019/790 (CDSM Directive)
3. Retsinformation, Ophavsretsloven (Danish Copyright Act, consolidated act no. 1093 of 2023)
4. Datatilsynet, AI guidance and decisions
5. Patent‑ og Varemærkestyrelsen (Danish Patent and Trademark Office, DKPTO)
6. WIPO, Artificial Intelligence and Intellectual Property