AI raises unresolved IP questions across copyright, patent, and trade secret law. Most jurisdictions do not grant copyright to purely AI-generated works (requiring human authorship), training on copyrighted data faces legal challenges under varying fair use and text-and-data-mining exceptions, and patent offices generally reject AI systems as named inventors while accepting AI-assisted inventions with human inventors.
Intellectual Property and AI: Copyright, Patents, and Trade Secrets
Copyright and AI Training Data
The legality of using copyrighted works to train AI models varies by jurisdiction. In the EU, the DSM Directive (2019/790) Article 4 provides a text-and-data mining (TDM) exception for any purpose, provided the rightsholder has not expressly reserved their rights in a machine-readable format. Article 3 provides a broader exception for research organizations. The EU AI Act (Article 53(1)(c)) requires GPAI providers to put in place a policy to comply with EU copyright law and to publish sufficiently detailed summaries of training data content. In the US, the fair use defense (17 U.S.C. Section 107) is being tested in ongoing litigation, with courts applying the four-factor test to AI training. Japan's Copyright Act Article 30-4 provides a broad exception for computational analysis not aimed at enjoying the work's expression.
Copyright in AI-Generated Outputs
The question of whether AI-generated content receives copyright protection turns on the authorship requirement. The US Copyright Office maintains that copyright requires human authorship, refusing registration for purely AI-generated works (Zarya of the Dawn partial registration, 2023; Thaler v. Perlmutter, 2023). The EU Copyright Directive requires works to be the author's "own intellectual creation" (Infopaq, C-5/08), which courts interpret as requiring human creative choices. The UK is an outlier: Section 9(3) of the CDPA 1988 assigns authorship of computer-generated works to "the person by whom the arrangements necessary for the creation of the work are undertaken."
| Jurisdiction | AI as Author? | TDM Exception for Training | Key Legal Basis |
|---|---|---|---|
| EU | No (human intellectual creation required) | Yes, with opt-out mechanism (Art. 4 DSM) | DSM Directive 2019/790, AI Act Art. 53 |
| US | No (human authorship required) | Fair use defense (contested) | 17 U.S.C. Section 107, Copyright Office guidance |
| UK | Possible (CDPA Section 9(3)) | Limited (Section 29A, non-commercial) | Copyright, Designs and Patents Act 1988 |
| Japan | No (creative expression required) | Broad (Art. 30-4, computational analysis) | Copyright Act, Cabinet Office AI guidelines |
| China | Emerging (some courts recognize AI outputs) | No specific exception | Copyright Law Art. 3, Beijing Internet Court rulings |
Patent Law and AI Inventorship
Patent offices worldwide have addressed whether an AI system can be named as an inventor. The EPO (J 8/20) and USPTO (Thaler v. Vidal, Fed. Cir. 2022) both held that an inventor must be a natural person. The UK Supreme Court (Thaler v. Comptroller-General, 2023) reached the same conclusion under the Patents Act 1977. However, all these jurisdictions accept patents on AI-assisted inventions where a human inventor made the inventive contribution using AI as a tool. The practical implication: document human involvement in the inventive process when seeking patent protection for AI-assisted innovations.
Trade Secret Protection for AI Models
AI model weights, training methodologies, and proprietary datasets can qualify as trade secrets under the EU Trade Secrets Directive (2016/943) and the US Defend Trade Secrets Act (18 U.S.C. Sections 1836-1839), provided they meet three conditions: the information has commercial value because it is secret, the holder has taken reasonable steps to keep it secret, and it is not generally known or readily ascertainable. The EU AI Act creates tension with trade secret protection: Article 13 requires transparency about high-risk AI systems, and Article 53 requires GPAI providers to make available training data summaries. Providers must balance disclosure obligations against trade secret preservation.
Open-Source AI and Licensing
Open-source AI model releases raise novel licensing questions. Traditional open-source licenses (MIT, Apache 2.0, GPL) were designed for source code, not model weights. Model-specific licenses have emerged (Llama 2 Community License, RAIL licenses) that impose use restrictions beyond traditional open-source terms. The EU AI Act Article 53(2) provides modified obligations for GPAI models released under open-source licenses, exempting them from certain transparency requirements unless they present systemic risk. Organizations using open-source AI models must verify that the license terms are compatible with their intended use and that the model's training data provenance does not create downstream IP infringement risk.
Practical IP Risk Mitigation
Organizations should implement an AI IP policy covering: (1) due diligence on training data provenance and licensing before model development; (2) contractual allocation of IP rights in AI outputs between providers, deployers, and end users; (3) documentation of human creative contributions to AI-assisted works to support copyright claims; (4) trade secret protection measures for proprietary models, including access controls, NDAs, and technical safeguards; (5) monitoring of AI outputs for potential infringement of third-party IP; and (6) review of open-source model licenses for use restrictions that may conflict with commercial deployment.
Check your AI compliance readiness — free.
Take the Readiness Check 3 minutes · 10 questions · no signup requiredThis article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.