As WIPO put it, until very recently innovation and creation was one of the defining characteristics of the human species. Machine learning has made that sentence require a qualifier — and IP systems built on the assumption behind it are still working out where to put it.
Where AI already sits
Films and television shaped our expectations of AI long before the technology arrived — HAL in 2001: A Space Odyssey, Samantha in Her, Chitti in Endhiran, G.One in Ra.One, most episodes of Black Mirror. The reality has been inching toward them.
AI now touches ordinary life at every point: phones, cars, washing machines, even insulin delivery. WIPO itself launched an AI-powered tool to help classify images under the Vienna Classification — the IP system using AI to administer IP.
WIPO’s response
WIPO led the analysis of AI’s impact on IP and the case for amending legal frameworks, establishing a dedicated forum — the WIPO Conversation on Intellectual Property and Artificial Intelligence — in 2019, bringing member states and stakeholders together to debate the trends and their consequences.
The position it identified: AI is useful for IP management and administration, while the protection of AI-generated output under existing frameworks is unclear.
The conflicts
AI is capable of being an innovator and creator — as the artworks, patents and musical works produced by AI systems demonstrate. That produces genuine dilemmas about ownership and infringement.
The impact on trademarks can be reasonably quantified. The impact of AI-created output on patents, designs and copyright is where the open questions sit:
- Where IP rights are tied to ownership, would an AI be the first rights holder of the design, work or invention it produces?
- Would the law recognise an AI as a legal person?
- Could an AI be sued for infringing another’s rights — or would the creator of the AI be vicariously liable?
The inventorship question was answered in the negative by the UK, US and European systems when the AI system DABUS was designated as inventor.
How that has settled. The refusals hardened into a consistent position. The UK Supreme Court confirmed in 2023 that an inventor must be a natural person; the US Federal Circuit reached the same conclusion in Thaler v. Vidal; and the EPO’s Boards of Appeal held likewise. The same reasoning governs copyright — the human authorship requirement upheld in the US in the Thaler copyright litigation, itself the principle established in the monkey selfie case.
So the answer across major jurisdictions is now clear: AI cannot be an inventor or an author. What remains genuinely unsettled is everything downstream — who owns AI-assisted output where a human contribution exists, how much human input suffices, and what disclosure is required.
The other side of it
The “Next Rembrandt” raised the question from the copyright direction: could the AI system hold copyright — or is the better question whether copyright is required at all for AI-generated work?
That reframing has aged well. If no one can own the output, it falls into the public domain, which is a coherent outcome rather than a gap. The commercial pressure to change that comes from those investing in the systems, not from any deficiency in the reasoning.
And the question that has since overtaken all of these concerns inputs rather than outputs: whether using copyrighted works to train a model infringes. That is now the central litigation in the field, and this post predates it.
The takeaways
- AI cannot be an inventor or author — settled across the UK, US and EPO.
- Ownership of AI-assisted output remains open where a human contributed.
- Liability likely rests with the creator or operator, not the system.
- Training data is now the live question — inputs, not outputs.
Frequently asked questions
Can an AI be named as an inventor on a patent? No — the UK Supreme Court, the US Federal Circuit and the EPO have all held that an inventor must be a natural person, in the DABUS litigation.
Can an AI own copyright in what it generates? No — copyright requires human authorship, so purely AI-generated output has no owner and falls outside protection.
Who is liable if an AI system infringes IP rights? The question is unresolved in India, but liability would ordinarily be sought from the creator or operator of the system rather than the system itself.
What is the main AI and copyright issue now? Whether using copyrighted works to train models infringes — a question about inputs rather than about the ownership of outputs.
Useful official resources
- WIPO — Artificial Intelligence and IP
- The Copyright Act, 1957