Attempt to get copyright on wholly AI-generated art gunned down by US Supreme Court

1 week ago 11

What means this for videogamez?

Ultron has his hand outstretched, countless other robots following behind him from a portal in Marvel Rivals. Image credit: NetEase

The highest legal authority in the USA has refused to take a case involving a Missouri computer scientist who wants to copyright art created by his own generative AI system. The compsci guy in question, Dr Stephen Thaler, appealed to the Supreme Court justices after lower courts upheld a US Copyright Office decision that his art doesn't make the cut for copyright protection, because it wasn't created by a human.

As reported by Reuters, Thaler's generative AI tech is the winningly dorkish "Device for the Autonomous Bootstrapping of Unified Sentience", aka DABUS. He applied for a US federal copyright registration, asking for DABUS to be recognised as the author of a picture of a train tunnel surrounded by mucky green and purple vegetation, called "A Recent Entrance to Paradise".

The Copyright Office rejected Thaler's application in 2022. A Washington judge upheld that decision in 2023, writing that human authorship is a "bedrock requirement of copyright", and the US Court of Appeals for the District of Columbia Circuit affirmed the ruling in 2025. The Supreme Court has also rejected a separate request from Thaler that AI-generated inventions – in this case, prototypes for a fancy food container and some kind of emergency light – should be eligible for US patent protection. Thaler's lawyers complain that if the Copyright Office's reign of terror continues unchecked, it "will have irreversibly and negatively impacted AI development and use in the creative industry during critically important years."

Thaler has been on the warpath trying to get DABUS-generated work through patent registration in many other countries, including the UK, European Union and Australia. So far, it appears he has only prevailed in South Africa.

The US Copyright Office have elsewhere shot down bids for copyright on images created with the 'assistance' of Midjourney. One company, Invoke, have successfully applied for US copyright on a piece of art, after demonstrating in detail that a human author “actively selected, coordinated, and arranged” AI-generated elements for the final composition. But that isn't what Thaler and his lawyers are after. They're applying for copyright and patents with the generative AI software itself recognised as the creator.

My knowledge of intellectual property law derives mostly from Ally McBeal and that one thing I wrote about Palworld, but if companies are able to copyright or patent the eerie fruits of AI generation, whether on 'the AI's behalf' or not, I imagine it'll be pretty consequential for the AI biz at large. The question of intellectual property is at the heart of the global argument around generative AI. Most generative AI software is produced by feeding large quantities of human-made Stuff into a program so that it can identify patterns, which are then used to stitch together outputs in response to a prompt.

Naturally, there is fierce interest among the genAI companies in getting access to new, proprietary or firewalled data sources for training, not least because generative AI models may "collapse" when trained on all the AI-generated bullshit that now swamps the internet.

Some generative AI companies explicitly license or otherwise own rights to the material they use for the dataset; many do not. I'm not sure what data DABUS uses, based on an hour's research. In legal documents, Thaler has strongly positioned it as an autonomous entity. With regard to the patent applications, he told a Swiss court that the model was not trained on any "special data relevant to the instant invention".

In general, if people who use generative AI – including game developers – have legal precedent for saying that outputs should be protected by copyright or patent law, it'll surely tip the balance of power towards the genAI companies in their efforts to demonstrate that they are making appropriate and acceptable use of other people's work for 'training'. DABUS might be a different kettle of fish to ChatGPT, but I can imagine a successful DABUS copyright application becoming precedent for Big AI's efforts to establish cultural legitimacy.

If I sound pissed off about this, it's because you can generate a wonky imitation of my own writing right now, using ChatGPT. It's not as bad as having your face spliced into pornography, but still, I feel that Sam Altman should have hand-written me a nice letter requesting permission, before handing over my mastery of split infinitives to the Torment Nexus. Full disclosure that this website's parent company, Ziff Davis, are currently suing OpenAI for "intentionally and relentlessly" misusing copyright works for chatbot 'training' purposes.

Read Entire Article