The Steam game store has a policy that games with AI-generated content are required to say so. The CEO of Epic Games (which owns a competing game store) recently criticized this policy:

The AI tag is relevant to art exhibits for authorship disclosure, and to digital content licensing marketplaces where buyers need to understand the rights situation. It makes no sense for game stores, where AI will be involved in nearly all future production.

This has been making the rounds as people react to it on social media. I thought I’d offer my two cents, since I just set up a Steam page a few months ago, and had to familiarize myself with the AI disclosure policy at the time.

To provide context on my perspective, I’m not an expert, I’m a first time hobby dev. For me, navigating the Steam backend is a bit overwhelming, because it’s obviously made to be used by professionals, not people doing it on the side for fun. I don’t stand to gain or lose much by commenting on this topic, because I’m projected to sell like ~20 copies either way. This, by the way, is not atypical, as research estimates that 40% of the games on Steam fail to recover the $100 fee required to create a Steam page. But obviously when people think about the typical video game, they’re thinking about the typical commercial game, and I’m not that.

Steam’s Content Survey

Steam’s AI disclosures are part of Steam’s “content survey”. When you create a Steam store page, you’re required to disclose whether the game has various types of violence, nudity, drugs, in-game purchases, and so on. The content survey is quite detailed, and it’s used to generate age ratings in various countries around the world. At the bottom there’s a section on generative AI, with a few distinct items. I’m not going to quote it directly since I think that’s against my NDA, but I can point to the public documentation.

The third part of the survey deals with generative artificial intelligence in your development or product.

If your game used AI services during development or incorporates AI services as a part of the product, this section will require you to describe that implementation in detail:

• Pre-Generated: Any kind of content (art/code/sound/etc) created with the help of AI tools during development. Under the Steam Distribution Agreement, you promise Valve that your game will not include illegal or infringing content, and that your game will be consistent with your marketing materials. In our prerelease review, we will evaluate the output of AI generated content in your game the same way we evaluate all non-AI content – including a check that your game meets those promises.
• Live-Generated: Any kind of content created with the help of AI tools while the game is running. In addition to following the same rules as Pre-Generated AI content, this comes with an additional requirement – in the Content Survey, you’ll need to tell us what kind of guardrails you’re putting on your AI to ensure it’s not generating illegal content.

As an aside, AI in games is rapidly evolving and we want to get it right. Many artists and creators are rightfully concerned about exploitation and misuse. But we also realize that technology and tools are going to change a bunch over the years, and there are lots of creative game developers out there figuring out ways to ethically use AI to make great entertainment. If you plan to release a product on Steam and you think we’re overlooking something important in regard to AI, we’d love to hear from you.

The content survey particularly emphasizes live-generated content rather than pre-generated content, basically expressing concern that a gamer will provoke a game to generate illegal content.  In other words, Steam’s primary motivation is to cover its own ass.  But in public, Steam instead emphasizes the AI disclosures as a consumer-friendly move.

Devs vs Gamers

Personally, as someone with no real financial stake in game dev, I am in favor of disclosing information. Some consumers don’t want any generative AI content, they’re like the vegans of the digital realm. Like with vegans, I don’t need to agree with them in order to support reasonable labeling.  I’m flexitarian, those labels are useful to me too.

But also like with vegans, there is a fear of the militant ones. In game dev spaces, I often see stories about negative reviews based solely on the perception of AI-generated content. It’s not clear how common it is, but it’s similar to hearing stories about games getting bad reviews because they have “woke” content. Game devs often feel this is unfair. For one thing, people often perceive AI-generated content where there is none. Other times, you know, it’s like you pour hundreds to thousands of hours into a passion project, but one thing you’re not passionate about is marketing, so you use AI to generate the Steam capsule. And then you get negative reviews accusing you of laziness, even though the AI image alone probably took an order of magnitude longer than the time spent by the reviewer.

I’m asking for a bit of sympathy for the game dev perspective, and in return I will grant sympathy to the reviewer perspective. Gamers generally under-appreciate how much work it takes to create even “shovelware” games, but it’s also not really a review’s role to give points for effort. If you don’t like the art or think it raises ethical concerns, that’s an opinion you can express. I also don’t really like most AI art, I’d really rather draw it myself, and its presence certainly doesn’t seem like a positive signal of a game’s quality.

In the case of commercial games, and especially large commercial games, I suspect the publishers are making a calculated decision about whether it’s riskier to disclose, or to avoid disclosure. When a game studio gets caught “accidentally” using AI images, I think it’s reasonable to suspect that they knew exactly what they were doing. For smaller games, motivations may be more personal.

I want to be honest and transparent about my game. It’s particularly important, because my game contains dialogue discussing generative AI! I did not set out to write about generative AI, but the protagonist is a robot, and in the year 2025 I feel like it only makes sense to write a robot in light of LLMs. But I did not use LLMs to generate any writing, and I want this to be clear upfront.  In my game, there’s no robot putting words into my mouth, I’m putting words into the robot’s mouth.

So I have to say, Steam’s policy on AI disclosures is not as clear as I would like it to be, and I’m concerned that it actively obfuscates issues.

Steam Policy and Practice

There are two things about Steam’s policy that stand out to me as unclear.

The second thing that stood out is, what kind of AI are we talking about? The policy starts out by talking about generative AI, which is a term referring to LLMs, image generators, music generators, and maybe some other stuff? However, it also says “content created with the help of AI tools during development”, which is much broader than generative AI! For example, I could create content with the help of Google. Search engines are unambiguously AI.  And if that weren’t enough, Google has been using LLMs like BERT to interpret search queries years before AI summaries existed.

Here’s a video talking about the many different ways AI is used in game development. For each application, you could ask whether it counts as “content created with the help of AI tools”. It’s legitimately unclear to me!

But obviously, no gamer is going to look at an AI tag and decide, “They must have used search engines to create this game”. No gamer is going to say, “Did you use search engines? Then where’s your AI tag?” That’s the point I’m making. There is a large gap between the text of Steam’s policy, and the way that any reasonable person would interpret an AI tag. Game devs are forced to apply their own interpretation to the policy, because the literal interpretation isn’t it.  And since they’re interpreting anyway, they’re incentivized to adopt narrow interpretations that will avoid a tag that could cause negative PR.

The first thing that stood out to me about Steam’s policy, is that it explicitly includes code. I think this is getting at something like vibe coding. Or agentic coding. Or AI autocomplete? Maybe it counts if you ask ChatGPT for coding advice? What about if you google a coding question and let your eyes fall upon the AI summary? What if you find an answer on Stack Exchange that someone else wrote with AI assistance? What if you import an open source package that uses AI to assist with reviewing PRs? Actually, I have no idea what’s covered under this policy.

AI assisted coding is a hot topic, and you’ll find a wide range of opinions even from programmers. I think there are legitimate concerns about how effective it is, and especially whether its effectiveness meets the optimistic expectations of certain investors. However, in order for AI assistance to be incorporated in games everywhere, the bar it must pass is extremely low! It needs only provide a marginal benefit in some area of development. And then, why wouldn’t you use it? Do gamers actually care about that stuff?  They historically haven’t given a shit about stuff like developer crunch.

I think that AI assisted code is far more common than AI-generated art or music. If we’re talking about code, Tim Sweeney is basically correct in saying “AI will be involved in nearly all future production.” So when it comes to Steam’s stated policy on AI disclosures, code really stands out as the thing that the policy is about. And yet when gamers talk about it, it’s all art art art art art and music. The gap between the text of Steam’s policy and the understanding of the typical gamer is huge. What the hell does a game dev do with that?

Since I’m interested in transparency, it doesn’t make any sense to strictly follow Steam’s stated policy. Because if I did, people would take the wrong message, because it’s not like they’ve carefully read Steam’s policy themselves, especially not the parts of the policy that are not public information.  I think the best way to decide whether a game deserves an AI tag is based on how it would be understood by the typical gamer. Which… who knows??  For my game, I didn’t give it an AI tag, and instead plainly explained that I used some AI assistance in coding but did not use it to generate art, music, or dialogue.