In-depth reporting and analytical commentary on artificial intelligence regulation. No legal advice.

Game distribution platform Steam aims to approve “vast majority of games that use [AI]” with new rules

Context: Steam (by Valve) is the most popular game launcher and distribution platform on PCs with a market share of approximately 75% (Wikipedia). In 2023, game developers started complaining about systematic rejections of games with AI-generated content (Reddit discussion board).

What’s new: Valve has now announced new rules (corporate blog post). Game makers submitting their titles for distribution via Steam must declare how they used AI and what “guardrails” they put in place against the potential infringement of third-party intellectual property and other violations of applicable laws. The disclosures will also be used to inform Steam’s customers. Valve’s blog post suggests those rules are merely a first attempt.

Direct impact: Valve will take those disclosures of AI use into account when reviewing games and now hopes to be able to approve “the vast majority of games that use [AI to generate content].” If Valve acted too restrictively, its competitors might get content that gamers would not find on Steam, potentially shifting market share. Essentially, Valve’s rules remind game developers of their liability for potential IP infringement or other kinds of violations. Game makers may seek similar warranties and representations, and possibly clear-cut indemnification clauses, from the providers of the content generation tools they use, though smaller developers will lack the leverage to negotiate individual agreements with service providers and tool vendors.

Wider ramifications: Valve’s description of AI technology as “legally murky” is unnecessarily negative. There are questions of first impression for many courts, and tatutes and case law are evolving. Valve’s blog post also misses the point that even without any use of AI, there is always a risk of IP infringement. Copyright law and (design) patent law are strict liability regimes, meaning there isn’t even an independent invention defense.

Valve deserves credit for trying to achieve a compromise between openness (toward games made with new content generation technologies or using such technologies in real time) and accountability with respect to potential IP infringement. The distinction between pre-generated and live-generated content is an important and objective one, given that pre-generated content is a finite set of material that could theoretically be reviewed, while live-generated content is potentially unlimited and not fully predictable.

The final sentence of Valve’s statement is also an honest admission:

“We’ll continue to learn from the games being submitted to Steam, and the legal progress around AI, and will revisit this decision when necessary.”

The corporate blog post refers generally to “illegal and infringing content.” While copyright comes to mind, content can also be illegal for other reasons (for instance, it could offend people), and even within the domain of intellectual property rights, there is not only copyright but there are also other design-related rights, such as U.S. design patents or registered European designs. While the breadth of a design patent depends on the claim language, it is generally a safe assumption that design patents can be asserted against material that does not infringe any copyright.

Either way, those intellectual property rights are strict-liability regimes, meaning that one infringes even through independent creation if someone else created (and, if necessary, registered) it before. In other words, copyright prohibits not only unauthorized copying in terms of “copy and paste.”

In order for AI to rule out infringement, it would have to compare the output against all third-party rights, which is simply not possible. Copyright is obtained automatically (though in the U.S. registration is required before enforcement), making it technically impossible to research all existing copyrights and keep clear of infringement. Registered rights (such as design patents) are researchable, at least to the extent the applications have been published (initially they may be in stealth mode), but then one might need Artificial General Intelligence (AGI) in order to form an opinion on whether a given design falls within the scope of a design patent (and on whether a pending application is likely to be granted).

The risk of inadvertent infringement is never zero under a strict-liability regime. It may be statistically very unlikely. For instance, this article is one of only a few discussing this topic, and is therefore highly unlikely to have so much in common, at the level of expression (as opposed to facts), with any other article on the topic as to infringe. But even music composers have sometimes thought they created a new melody, only to find someone had already done so before them.

Furthermore, if game makers download content from an online store that does not allow AI-generated content, there still is no assurance that the creators of those packages didn’t infringe manually or that they didn’t provide untruthful assurances of not having used AI tools. The same applies to “gig economy” projects. For instance, the logo of this website was created by a designer on fiverr, who apparently didn’t even create the logo himself but used third parties. No one buying artwork anywhere on the internet these days can be 100% sure that no AI tools were used.

The search for solutions has merely begun.

Finally, credit to a copyright holder—the AI logo in the image above this article is from juicy_fish on freepik—and an editor’s note:

This is the first article (and there won’t be too many going forward) that cuts across all three fray sites: it’s about AI-generated games content raising legal issues, particularly with a view to IP infringement. What will happen more often is that a story is relevant to two fray sites (for example, an IP enforcement action against an AI service provider or against a games company). Whenever a topic is of interest to two or all three fray sites, an editorial decision will be made to which one to assign it (based on an assumption as to where there will over time be more stories discussing overlapping issues). If a fray site could have reported on a topic as well, it will share the sibiling site’s article on social media, and in exceptional situations there may be summaries or periodic overviews that appear on one fray site and link to in-depth coverage on another. Here, given the enormous market power of Steam in game distribution, games fray will publish a short item pointing to this article.