A Failure of Platform Policy: The AI Shovelware Problem | 23/07/25
20% of games published to Steam in 2025 have AI disclosures.
The AI and Games Newsletter brings concise and informative discussion on artificial intelligence for video games each and every week. Plus summarising all of our content released across various channels, like our YouTube videos and in-person events like the AI and Games Conference.
You can subscribe to and support AI and Games, with weekly editions appearing in your inbox. If you'd like to work with us on your own games projects, please check out our consulting services. To sponsor, please visit the dedicated sponsorship page.
Hello all, welcome to this week’s edition of the
newsletter. This week, I had originally planned on reflecting on my predictions for 2025 as a sort of mid-year reset. However, I realised that my final (as yet unrealised) prediction, is the source of the broader conversation right now on AI-laden shovelware. So this issue, I kinda take a stab of both of these things. So yeah, grumpy Scotsman activate!Follow AI and Games on: BlueSky | YouTube | LinkedIn | TikTok
Announcements
Just a couple of quick announcements around all things AI and Games!
AI and Games Conference 2025: Speaker Submissions Close Next Week
That’s right, the speaker submissions for this years conference are winding down next week, and the review phase begins! Submissions close on August 1st, so be sure to head over to the website and get them in. Critically, a little warning that the review form is longer and more substantial this year. It will take you at least 15-20 minutes to fill it in. So if you’re planning on putting in a submission, make a cup of tea or coffee and carve out some time!
For our final inspirational quote, we have the final member of the Game AI Events CIC board, Alan Zucconi from our host Goldsmiths University. Alan speaks of passion here in his quote, and he’s someone who lives by that principle. Every student I’ve met who went through the Goldsmiths course to this day speaks of his passion for teaching game development and driving them to achieve their best. The same applies of course, for AI and Games Conference speakers!
AI and Games @ Gamescom 2025
We’re knee deep in AI and Games Conference prep this week. So no new updates on Gamescom yet, but when we have something to announce, we’ll let you know.
Keynote Speaker @IEEE CoG 2025
I’m pleased to announce that the week after Gamescom, I’m returning to my academic roots by attending the IEEE Conference on Games 2025 at the Instituto Superior Técnico as part of the University of Lisbon.
Since 2019 CoG has been the Institute of Electrical and Electronic Engineers’ (IEEE) premiere venue for the publication of research in games. However, CoG’s predecessor was the IEEE Conference on Computational Intelligence on Games (CIG), and that’s where I published my first ever academic publications on AI in games all the way back in 2007. So to celebrate 20 years since the inaugural CIG symposium (hosted at the University of Essex in April of 2005), I’ll be delivering a keynote looking back on how games research in and around games has evolved. Here’s the blurb:
Days of Future Past: Celebrating 20 Years of IEEE CoG
In 2025 the intersection between artificial intelligence and video games has never been stronger. Frontier AI research frequently uses games as a platform, meanwhile games companies face the new challenges presented to them in an age of generative models, plus the still untapped potential of conventional machine learning. But in 2005, it was a very different story: of ‘simulations’ that hid their ludic origins, where machine learning was the sole domain of academics, and these explorations seldom aligned with corporate realities. To celebrate 20 years since the inaugaural IEEE Symposium on Computational Intelligence and Games, we look back on how research in AI for games has evolved, how the industry evolved alongside it, and the challenges that face both sides of the coin in an era of AI proliferation.
A Failure of Platform Policy
Alrighty, time for a big write-up of the week. At the start of the year I made a bunch of predictions about how AI would continue to impact the games industry. And I figured why not start this issue with a reflection on this topic. Though to spoil the ending a little, I’m both happy and disappointed to report that all bar one of those predictions became true within six months. The final point is now something that games platform holders have been slow to act on, and it’s creating a broader conversation about the quality of said storefronts and the permissiveness of generative AI in game releases.
Reflections on Predictions
So yeah, back in January I sat down and had thoughts about how and where AI in the games industry was going to go. I had tried both to refrain from buying into much of the hype around where things are going, as well as refrain from stewing in my own cynicism (which was the harder part believe me). You can read the article in full, but I’m going to quickly summarise each point:
A Make of Break Year for AAA Gaming:
I think it’s fair to say that the larger big game space is having an interesting year. While we’ve had a whole bunch of well-performing and critically received titles this year, ranging from Avowed to Monster Hunter: Wilds, Death Stranding 2, to DOOM: The Dark Ages, Assassin’s Creed Shadows, and Split Fiction, there are still many a title that has struggled to carve out big success while others have bombed at launch. Despite being a very good game, Civilisation VII is facing an uphill battle to get players on board, while Mindseye is arguably the big bomb of the season thus far - well, I guess Splitgate 2 heard me writing this paragraph and decided to respond. But all of this sits at the front of an industry where finding success continues to be a challenge, all the while gathering the necessary funding and talent becomes all the more tricky.Layoffs Will Continue into 2025
According to the VG Layoff tracker, we’re at around 4,000 confirmed for 2025 - with not all of the recent layoffs at Microsoft confirmed - or all based at Xbox studios. But still, I suspect the number is at least 20% higher in reality.NVIDIA’s Push Into Full-Fledged AI Company
So this prediction was largely proven true as I wrote the issue. Go back and read where I talked about the highlights of CES 2025 where NVIDIA more or less proved my point - more AI-focussed hardware, neural rendering and expanding the ACE platform for generative AI focussed execution in games and related technologies.More AI Demos, More Start-ups, More Local Inferencing
So many start-ups that gathered up millions of dollars in funding are still cooking. At GDC we saw the likes of BitPart.ai talking about their agent architecture that allows for multiple characters to execute in a scene - which struck me as a good step in the right direction. I mean just last week Artificial Agency, who we reported on in July last year when they pulled around $16 million of funding just starting making the rounds with early demos of their ‘behaviour engine’.Meanwhile games such as Dead Meat made the rounds at GDC earlier this year with their talk at the NVIDIA stage where they talked specifically about how they have built a version of the game with local inference capabilities, meaning the large language models (LLMs) that they use for conversation and story can execute on the same device as the game, and not require an online connection.
AI-Native Games Will Start to Ship, But Mostly From Indie Studios
A reminder that ‘AI-Native’ is a term that means the use of generative AI as part of the core gameplay experience - it’s a reductive term, I don’t like it, but hey, it’s what the kids are saying. But yes games like AI2U: With You ‘Til the End, and Millenium Whisper are on Steam as games that use generative AI as part of the core gameplay. Meanwhile InZoi is arguably the most high profile game of this nature, which as I discussed just recently appears to have lost the interest of most players for now.Someone, Somewhere, Is Going to Put Their Foot in It
I mean, I still have a rant brewing in me about that whole Darth Vader thing. But don’t forget the whitewashing episode that NVIDIA created with their neural face rendering technology.Platform Holders Will Begin to Dictate Policy
And here’s where it gets interesting, because this is the one thing that hasn’t happened (yet), and it’s beginning to become a problem.
Building AI Policy for Platforms
So yes the final point I raised in my January issue that has not come to pass is that I argued platform holders need to take a stronger stance - well, any stance - on the adoption of generative AI technologies on games released via their premium storefronts.
Now of course it’s worth raising what I mean by a ‘premium’ storefront: Steam to some extent, but really what I mean are the PlayStation and Xbox stores, and the Nintendo eShop. I phrase it this way because mobile stores such as the Google Play and Apple App Stores have been flooded with questionable content for many years, and the reason I focussed on the consoles (and to a lesser extent Steam) is what we consider ‘premium’ games experiences, whether they be developed by indie or AAA studios, are those that come out on consoles.
I argued in January that this was important both to further reinforce and build upon that which Valve had already done with its AI disclosures, but also that it was in console manufacturers best interest to anticipate the backlash and negative sentiment that generative AI would bring to the gaming community.
No Game is Bigger Than the Platform
All three of the main platform holders have gone out their way the 10-15 years to build initiatives that support indepedent developers in getting their games onto consoles - PlayStation Indies, Nindies, and ID@Xbox - so that they can support them in handling the challenges of shipping to console. This is something of a necessity because - speaking from experience - shipping games on consoles is a lot of work. It’s not just porting the game to run on the hardware, it’s also about integrating platform features to work in your title, it’s about respecting the guidelines of the target hardware to ensure you follow their rules, it’s about satisfying store requirements so ratings are correct and you have information presented in various markets. Getting all this done on Steam is a lot of work as it is, getting it all into place on console is arguably harder, and it’s a key requirement for shipping.
This all comes to a head when you enter Lotcheck: the process of submitting your game to review. Those aforementioned guidelines? You need to satisfy them, and almost no game ever succeeds lotcheck on the first attempt. Because in order for a game to ship, it must satisfy the rules of platform in a way that ensures a quality experience.
No game is bigger than the platform, and no game should bring the quality of the platform into disrepute.
This is what led to the likes of the digital edition of Cyberpunk 2077 being pulled from the PlayStation Store back in late 2020 - at a time where it was riddled with bugs and performance issues - because it raised questions of why Sony permitted the game to be released in the state it was for a hefty price tag.
Individual Failures, Collective Negligence
One of the earliest issues of this newsletter discussed the impact of Valve updating and clarified their stance on generative AI. At the time I highlighted this is an important step to mitigate another wave of low-quality titles flooding the Steam storefront. Steam is notorious for ‘asset flips’ and shovelware appearing on the store, with plenty of low quality titles flooding the market and making it harder for smaller developers to stick out in the crowd.
Meanwhile, console storefronts needed to start clamping down on this as well. All of the console holders have a growing problem of low-quality titles bleeding into their storefronts. The Nintendo eShop back in 2020 struggled with poor quality games that adopted aggressive discount strategies as means to float to the top of their sales charts. Hence it was important to anticipate the influx of titles that generative AI would bring, and in turn how to combat low quality and spurious submissions to retain the value of those storefronts.
Valve’s Negligence is Historically Consistent
When I wrote that article in early 2024, the data showed that 14,000 games had shipped to Steam in 2023, though only ~4000 of them qualified out of SteamDB’s ‘limited games’ bracket: a term used to denote games that fail to achieve a certain threshold of community customisation features - which in turn often implies the overall commercial success of the game. Of that 14,000, only 117 games provided AI disclosures. That’s a meagre 0.8% of all games in 2023.
However, I equally argued that these rules were not explicit enough. It needed to be clear what is deemed acceptable, and there needed to be further controls on flagging this content - be it to filter it out of search results, or report broken features. Combine this with Valve’s historically low bar for content moderation when games ship on the platform, I wasn’t confident that this would stymy the flow of low-quality games that ship onto the platform that now exploit generative AI as a means to an end.
So it came as no surprise to me when Totally Human recently reported the surge of Steam games that utilise generative AI tools as part of their projects. Combined with the data from SteamDB now showing that over 2000 games have signed the AI disclosure in 2024, and already at the time of writing over 2000 have disclosed AI tools in 2025. To put that into context, there were 18,700 games in 2024, and already we’re at 10,500 for 2025, so that’s 10.7% in 2024, and 19% in 2025.
While these disclosures were made as part of the process, there’s seemingly zero effort made to manage and curtail them:
Low quality titles are not barred.
Generative arts that appear highly derivative are not removed.
Some games that are clearly riffing on other titles get through the process.
Games that fail to disclose generative AI do not appear to suffer consequences.
There’s no means to filter out games with generative AI in the Steam store.
A crucial point I will return to shortly.
Meanwhile on Console Storefronts…
All three of the major platforms have had similar issues with more games that use generative AI. More recently Zack Zwiezen over on Kotaku reported about the influx of AI shovelware on the PlayStation store. Meanwhile with the launch of the Switch 2, Nintendo made another change to the eShop to essentially prevent low-quality (often AI-heavy) titles bubbling up to the top of their store charts.
In each case, it comes back to a fundamental problem of quality control. Many of the games cropping up in these instances are staggeringly poor quality, and largely flies in the face of the supposed quality processes that indie games have historically fought through to be released on those platforms.
What’s Generative AI Being Used For?
It’s important to take a moment to highlight both what generative AI is being used for in these games, plus the distinction that not all games that use generative AI are inherently low-quality products. My argument here is that while some studios will utilise the technology for very specific purposes (see The Finals or Liar’s Bar for AI voices), there is equally a need to distinguish those from what amounts to poor quality titles that are clearly exploiting AI tools to make up for a lack of agency and authorship.
Per Ichiro Lambe’s reporting on Totally Human, it seems that most generative AI adoption falls into the following categories:
Asset Generation: Art, textures, 2D and 3D models.
Audio: Music but also voices for characters be they fully generated in advance, or through use of speech-to-text systems.
Text: Story beats, item descriptions, and even entire narrative generation.
Code: Unsurprisingly, a lot of people are using tools like GitHub copilot.
Marketing: A lot of store pages, and their promotional art, is AI generated.
Almost all of this falls into the space of using AI tooling to offset the costs of specific parts of development, be it character and concept artists, musicians and voice artists, and even marketing teams. It very much falls in line with existing shovelware and asset flip behaviour, in that people exploit mechanisms to get this material as cheaply as possible rather than going the route of working with professionals.
Perhaps the biggest issue here is that Valve’s guidelines are so vague that’s it is difficult to really capture any meaningful understanding of how and where generative AI is used. Many a title out there uses very soft language to hide just how pervasive the AI is. Plus of course, recent events such as The Alters highlights that no doubt many studios are using it and keeping it hidden.
As such, the totals being reported everywhere are no doubt simply the lower bound on a much larger figure.
Customers Are Savvy
I’ve said it before and I’ll say it again, that using generative AI at its lowest, most exploitative form, of generating assets (often using third party tools that are legally ambiguous at this time) is not a good look in the eyes of many players. So many games are being caught out during or after development because of this, and there’s a big question for any studio to ask as to whether this is important or valuable to them in their creative processes, given it could have lasting consequences or impact the conversation surrounding your latest game.
The ‘No GenAI’ movement is strong not just among creatives, but also among players. The well has been so thoroughly poisoned by the exploitative uses of the technology that many people simply won’t give your game the time of day. Steam curators are popping up to report any game they find with generative AI content - regardless of the actual quality of the product.
While I still believe that ‘AI-Native’ games should have their chance to prove themselves, there is equally a need to recognise the broader impact of this significant increase of content hit these platforms. At a time where people are increasingly sceptical of spending money on games, the market is being flooded with more and more poor quality efforts. It’s no surprise that when SteamDB created a filter for AI content it was met with a lot of press and much praise among gaming communities.
It’s becoming increasingly clear that a lot of consumers arguably don’t want this, and will take steps to avoid it where possible.
The Value of the Platform
Despite my framing it as such earlier in this article, at some point in the past decade, all of the major storefronts lost sight of the perceived value their editorial constraints presented. That by curating the content that emerges, and mitigating the shovelware, it acts a sign of quality that the games that make it onto those storefronts are worthwhile of your time and investment - whether you actually like the games is a matter of personal preference of course.
Of course Valve’s approach to Steam has largely been to put 99% of the content submitted to them online, given at the end of the day the money is going in their pocket. Be it the Steam Direct fee, or the handful of sales that any one title achieves. Steam has had problems now for about a decade of smaller titles struggling to cut through the noise despite being decent games in their own right. Sometimes it’s only when they reach console players - who themselves are a smaller but more mainstream consumer audience - that many of these games find success.
Given this ‘filtering’ process, it’s a sad state of affairs that none of the console owners have thought about retaining that value. Trying to get a game through ID@Xbox 10 years ago was really hard - in fact when my friends and I shipped our own wee game back in 2018, it was initially rejected. Now the bar seems incredibly low for games to make it onto the platform. Though I suspect nowadays Xbox is going to be the last storefront to in any way demonise generative AI adoption given how all-in Microsoft is on the topic.
All of this has a negative impact on discoverability when consumers begin to simply ignore the upcoming or new releases pages of storefronts. Grabbing that screenshot above of new releases on Xbox was the first time I’ve opened that page in years, because I just glaze over when looking at it. For every game that could be of interest, there’s five where I question how it got up there. All at a time when it’s harder than ever for developers to get people to pay attention to the games that they’re making.
This is ultimately a reminder that generative AI is increasingly more prevalent on storefronts, and for every well-intentioned developer - and even interesting or fun game - it acts as a catalyst for opportunistic shovelware merchants. It’s a a separate axis of enshittification, as now markets are flooded with more low quality efforts, and that makes surviving in this space as a small creator all the more challenging.
Wrapping Up
So there we have it, that was something of a mish-mash of a ‘mid-year reflections’ issue, while also focussing on a current topic. Not bad huh? I was a little more cynical than usual this week as well. Sorry about that, more unfiltered optimism in the future!
On a slightly upbeat note however, the recent trailer for Mafia: The Old Country looks pretty damn neat. Some good ol’fashioned game AI right there. Really interesting stuff. I’ve never really paid much attention to the franchise up until now…
Alrighty, I’m on the road again next week - but this time for a brief family break - but we will have the sponsor newsletter next week where I’ll be sharing with our premium subscribers a bunch of updates on what’s to come.
Thanks for reading, and catch you all next week.