Microsoft’s New AI-Powered Patents Offer Little to Be Excited About | 30/09/26

Hello one and all, and welcome to AI and Games. This week we’ve gotta whole bunch of conference announcements from both 2025 and 2026, plus a game review and a deep-dive into a new story from last week about AI patents submitted by Microsoft that were recently published.

Before we get into the weeds on everything else, a quick plug for last week’s AI and Games Podcast episode with friend of the show Andrei Muratov. We had a great conversation at Gamescom about the state of AI exploration in the industry right now and how this affects Andrei’s work at AWS. Worth checking out for sure if you haven’t already.

Follow AI and Games on: BlueSky | YouTube | LinkedIn | TikTok

AI and Games Conference

We’re around 7 weeks out from our 2026 event, but before that here’s some stuff from 2025! Ticket sales have been flying since our first announcements last week and we’re now expecting to welcome a few hundred of you with plenty more to sign up in the coming weeks.

More 2025 Talks Live on YouTube!

We’re still chipping away on a handful of talks from the 2025 event that haven’t made their way to YouTube. Once again my apologies, this is on me given I make the final pass on all the videos, and my schedule has prevented me from being able to do so!

So with that in mind, what have we got?

On‑Device Generative Agents for Lifelike NPCs
Piero Molino / Studio Atelico

With all the hype surrounding the opportunities to showcase LLMs in behavioural contexts for games, you’d be forgiven for forgetting most of this is impractical in traditional video game platforms. This leaves devs at the mercy of slow and costly online infrastructure. In this talk Piero Molino from Studio Atelico showcases their efforts to build proprietary LLM tech stacks designed to run on consumer hardware with zero inference costs.

Build Your Own Navigation Stack!
Bram Ridder / Kythera

Navigation meshes are now a standard shipped into most commercial video game engines. However, the features you have access to are at the mercy of the engine provider and how much access you have to manipulating the navmesh systems. With that in mind Bram Ridder from Kythera gives a breakdown of how to build and modularise traditional Recast navmesh tools to allow for more modular flexibility.

Heat, MaxTac, and Blockades: Expanding the Police System in Cyberpunk 2077
Maciej Celmer / CD Projekt RED

One of the biggest - and loudest - presentations from 2025 was CD Projekt RED visiting us to share their work in updating the police systems as part of the big 2.0 update for Cyberpunk 2077. Maciej crushed it in this talk. It’s a lot of fun, and I hope you enjoy it!

More 2026 Talks Announced!

Last but not least we have three more talks at this year’s event to announce. Having announced a whole stack of talks last week, we have a nice spread of talks to share with you today!

Giving NPCs a Voice: Expressing Behaviour Through Barks
Philip Dunstan / Katapult Ridge

I’m excited to welcome Phil Dunstan, former AI lead on games such as Tom Clancy’s The Division 2 and Star Wars: Outlaws to give his perspective on how to make great game AI. After delivering a great presentation in our track at Gamescom Dev back in August he’ll be taking the stage to share his perspective on how to use simply bark systems to maximise gameplay and communicating to players quickly what is happening in the moment.

Your LLM Badly Needs a DM: How Player Research Taught us the Perils of Unlimited Freedom
Richard Cole / University of Bristol | Thomas Keane / Meaning Machine

At the very end of their presentation at the AI and Games Conference 2025, Thomas Keane and Ben Ackland shared that they were collaborating with the University of Bristol conducting player research into utilisation and engagement with LLM-driven systems. So this year we’re getting an update on how that research is progressing.

QA Without Playing the Game: Using LLMs to Generate Solvers for Game Rules
Joris Van De Donk / AWS

Last but not least for this week we welcome Joris Van de Donk from Amazon Web Services who will be presenting on a rather interesting concept: of using LLMs to help re-assert the rules of a game as it’s being designed by examining the source code. The idea being that it can help mitigate progression bugs in games by identifying when the game is currently in a condition that it can create situations where players should not be able to reach, or situations you cannot recover from.

As a reminder, the event is running 10th/11th November at HereEast in London.
Grab your tickets now!

Grab Your Tickets

Review: Kusan - City of Wolves

CircleFromDOT / Steam, Nintendo Switch, PlayStation, Xbox Series
Reviewed on the Nintendo Switch 2 with a review code from the publisher.

In recent travels I’ve been slowly chipping away at Kusan: City of Wolves, a top-down action combat game that is succinctly described as ‘Hotline Miami for furry lovers’.

I mean this might sound reductive but the game is very much an homage to Dennaton’s top-down shooter from 2012. Developed by South Korean developer CIRCLEfromDOT Player’s take charge of Jin, an anthropomorphised cat that wields both a cybernetic arm and a healthy bout of PTSD, who descends into a nightmare scenario after a job gone wrong. With an increasingly complex and hard to follow storyline involving criminal syndicates, civil wars, psionic weaponry, and sexualised animals that I’ll concede lost my interest almost from the beginning. It introduces more lore and backstory than the game warrants, and your mileage will most assuredly vary depending on your willingness to follow along, but I started skipping cutscenes pretty quickly even though they were rendered in a nice comic book style.

However the game really shines in its gameplay, and my remark that this is an home to Hotline Miami is immediately evident upon seeing it in action be it in the trailer above or in screenshots. The top-down perspective, the nice pixel-driven art style, the intense gunplay, the emphasis on perfecting each and every run. All of those core principles from Hotline Miami are either lavishly reproduced, or reframed and reintroduced throughout Kusan. With the ever present ranking system designed to make you think harder about how to maximise the utility that each enemy presents in an effort to maximise score.

Despite this, the game does have many of its own ideas that build on Hotline Miami such that it isn’t just a carbon copy. The game starts at a much slower pace, with Jin moving slower than you might expect, and has no capacity to use or collect firearms. The game then has you learn two of its key mechanics early on: the augmented arm and the knife. The arm allows for melee attacks, but also for the player to smash through doors and walls, and aim the debris at enemies on the other side to catch them off guard. But perhaps the most useful thing with the arm is that a well-timed punch counters a gunshot, and works as a parry technique. Meanwhile the knife is a ranged attack weapon that can either kill an enemy or disable them long enough for you to walk over, punch em out, and retrieve your blade.

The parry system feels like the hidden ingredient the developers want you to use when fighting against waves of enemies. You eventually unlock ranged weapons of varying sorts, but ammunition is limited and you cannot pickup weapons from downed foes. So when you’re in a pinch you should be able to save yourself with the parry and help it maintain your combo meter for the all important rankings. That said, I never felt like a had a grasp of the the timings for it, and seldom used it given I never felt confident with it.

The other big innovation is it has a fairly deep upgrade system, with currency collected in each level as well as through endgame rankings can then be employed to unlock new features for existing moves, or unlock passive upgrades. Upgrades are effectively chips that you install on a grid-shaped panel, and you have to maximise utulity of that space.

The main campaign has a rather varied setup, with different levels taking place behind closed doors or on the streets. With varying NPC types with different behaviour patterns, and even bosses in some situations that introduce a new challenge as you seek to work through a whole sequence without taking any damage. This is both a positive and a minus as some levels get increasingly complex and can begin to prove frustrating. I felt this was both a reflection of its camera being too low for the level design the game has built, combined with the reward structures forcing you to keep moving forward to get anything beyond a terrible rating, only to hit the wrong target at a key moment.

This brings me to what is perhaps my largest frustration with Kusan: the controls. This was both a reflection of the game and the platform. The basic movement can feel a little stiff at times, and the auto-targeting system can prove inconsistent as it shifts from one target to the next just as you set your sights on your next victim. This can be a real pain in the ass when you think you’ve got something lined up and then it flips out at the last moment. Secondly the bulk of all of the action are executed courtesy of the shoulder (LB/RB) and trigger (LT/RT) buttons which on a Nintendo Switch is not ideal. I had to dock the device and play on a Pro Controller in order to work my through some sequences. That’s not the game’s fault, but a word of warning for any Switch users: your hands may well cramp up.

Overall, for people that love the intensity of Hotline Miami this is guaranteed to prove of interest. When the game clicks and you get that near perfect run it feels very satisfying. I enjoyed the various scenarios the game puts you in, but not sufficiently for me to stick it out and aim for the highest of scores in each sequence out there.

• Fast-paced and intense combat that matches its inspirations.

• Quick and seamless replay, with more to have you coming back.

• Varied encounter design and NPC types.

• Interesting upgrade systems that force you to plan out resource spending.

• Narrative is a distraction and not really relevant.

• Controls feel inconsistent and stiff.

• The game can prove frustrating rather than fun in its most intense segments.

Thanks for reading AI and Games! This post is public so feel free to share it.

Share

Microsoft’s New AI-Powered Gaming Patents

Last week a couple of different news stories cropped up exploring new patents submitted by Microsoft that explored AI intersecting into games, and I figured we should check them out. We keep seeing weird and wonderful AI patents crop up, such as the ‘AI for Gaming Remaster’ patent from Sony, and Nvidia’s failed attempt to patent an AI-driven hint system in 2025 and earlier this year we read up on how Sony patented the idea of an ‘AI-driven gaming podcast’ that frankly sounded horrid.

Sadly neither of the patents we will cover today will get you any more excited. Two patents were submitted by Microsoft to the US Patent Office (USPTO) earlier this year:

• Contextually Aware Management of Interactive Software Application Access:

  • USPTO#: 20260260258 / Published September 3rd 2026 [PDF]

  • TLDR; a system that is designed to learn when best to present in-game ads to users. Odds are it will have to use some form of AI to achieve that.
    

• Generative Narrative Game Experience with Player Feedback

  • USPTO#: 20260183672 / Published July 2nd 2026 [PDF]

  • TLDR; train an AI system that can take on player feedback to change and alter the behaviour of a game, notably its narrative, based on player feedback.

The first sounds horrid - and unexpected - and the second sounds like a rather naïve assumption of how AI models can interface in traditional video games. Let’s look a little further.

What All Players Want: Dynamics Ads

Okay, so in the first patent the ambition is to find the best time during a game that is being streamed via a cloud platform to provide a “promoted content trigger event” can occur. In other words the idea that the game can have some form of advertisement presented while you’re playing. As we can see from the diagram above, the idea is that the ‘interactive software application’ (your game) is hooked into an ecosystem courtesy of the ‘content management system’, which is essentially where your game is being streamed from. The system is then awaiting one of these promoted trigger events which will allow for ‘promoted content’ to be pulled down from the corresponding server and the delivered to the user.

Now the key part is when do these events occur? Now it does appear to be time-based, and there’s a ‘credit’ system people can earn that will help them offset the next advert, but the once the user enters the ‘promoted content window’ then it’s waiting for the right time to inject the advert and suspending the user’s ability to interact with the game in the process. As the patent states some of these trigger will emerge from the software itself: it can send messages to the server denoting for example that the game is loading, or it’s paused, or that a specific thing has happened like the end of a round of Call of Duty - though that would need to be developed into the game itself. But the key thing for us is that they’re also looking at how to use an AI model to try and identify when to interrupt so that “the user’s experience may be improved”, or rather the user will be less pissed off when you interrupt at a key moment.

Several pages of the patent discuss the importance of using an “ML model” - they mean an artificial neural network judging by the diagram - that will help identify when not to put in an advert. Given it will recognise key moments of a game or the build-up towards them that would prove detrimental should they go ahead with it. For example, the patent lists moment-to-moment situations in sports games, FPS’ and MOBA’s where interrupting the game would be very negative. So capturing information of how people play a game in the cloud but also what is happening in those games is useful for it to know when not to interrupt.

I can’t help but notice this is also a convenient mechanism for Microsoft to get a whole bunch of free data on how people play Xbox games which I can imagine would be used for a lot more than just advertising!

Now none of this should sound surprising given the ongoing efforts by Microsoft to cannibalise what little value still remains in Xbox Game Pass. Recently it was announced that the cloud offerings would now incur usage limits on top of the base subscription, and players who wanted to play these games further will have to pay more. This apparently will only affect 4% of all Game Pass users, which is admittedly a much higher number than I expected.

They also announced that starting in November you will be able to stream Xbox games without a Game Pass subscription. Given this leaning into their cloud offerings - notably because their hardware is now so expensive it’s arguably not worth it - it’s no surprise that an advertising layer is being considered. They already ran a pilot for this that was open to Xbox Insiders, but that had much more static placement. This appears to be an active effort to start injecting advertisements based on the games you’re playing and the way in which you play them.

Re-writing the Story and More With AI

The second patent is for an idea where AI foundation models (i.e. GPT) are used to “create new game content responsive to player input”. The idea being that generative models then act to “aid in the creation of the game narrative” for game designers, while also give players “the potential to become co-creators of the game” by tracking their behaviour and even their feedback. Then generative models can be used to create new content. There are a series of AI agents running in the backend that will then create new game content that is responsive to the player’s input on the fly. The idea then being that either designers can review and potentially integrate these AI-generated suggestions into the game for future playthroughs, or an AI just does that itself.

A lot of this is predicated on the ‘engagement metrics’, and how collecting information from players can then inform or influence future creative choices. Now I have to say that hidden in amongst this sounds like something that almost makes sense. Let’s take a quote directly from the patent application:

The engagement metrics may be rule-based and defined based on heuristics. For example, the length of a player's conversation with an NPC or interaction with an object (e.g., measured in the number of dialog turns or individual actions or manipulations), the number of times player returns to a given location or asset, the completion or abandonment of a quest, and the selection of a path among multiple options at a branchpoint in the storyline may all be implicit indicators of the player's engagement with different parts of the game. Such rules may also analyze the intensity of an interaction; for instance, in a fight, a successive selection of ever more powerful weapons may signify an escalation indicative of high player engagement. Beyond the engagement data implicit in the player's interactions with the game, engagement metrics may also utilize features indicative of real-world reactions of the player to the game. For instance, audio and video signals, e.g., acquired with a microphone or camera, may be processed (with player permission), e.g. with the help of locally executing machine learning models, to infer such indicators from the player's spoken comments, facial expressions, etc. during gameplay. Self-recordings or live streams posted online by gamers may, for example, be processed in this manner. The derived indicators of real-world reactions can then flow into the computation of the engagement metrics 120.
US 2026/0183672A1 Page 3.

Now this part at least sounds interesting, in that it sounds like a mechanism to accumulate data from focus testing and absorb it into a system that allows for summaries of these datapoints to be expressed to the developer. This sounds like it could prove useful in running user studies, of having groups of people come in and play your game and have systems accumulate these engagement metrics and produce insight into what was found. Now this is something that already is conducted to some degree in user testing, and even relies on machine learning to do it when running analysis of the data. So having this accumulated into generative models that can speed up that process may well prove of value to developers.

But the thing I’m less enthusiastic about is the notion of modifying narrative graphs based on user input. Now we know from a variety of RPGs and the like that narrative designers build games around key events happening in sequence, and that the player will arrive at one of these events from a number of pre-determined paths. Now the patent suggests the idea of monitoring how players interact with a game and from that modify the narrative state graph to suit.

In theory I can see something useful here in working around the tight confines of narrative graphs, but as described it’s more about how players and the metrics used to report on them would be adopted as mechanisms to influence how the narrative graphs are changed. Now at least there is the opportunity for a designer to have a hand in that, but it seems more likely that it is about how a generative model could drive it.

I’ve actually had conversations before with narrative designers about the opportunity for generative models to sit in the middle of these graphs. Of using generative models to figure out logic that is baked in the ontology and rules of the game world that would help us guide players into narrative graphs more seamlessly. So rather than edit the graph itself, find ways to get players into specific states of the graph without violating the narrative space itself.

Wrapping Up

Well, all-in-all none of that sounds fun whatsoever. The first one obviously, I mean who wants commercials when you’re playing a game? But then I suspect this is all part of their efforts to sell Xbox as a cloud gaming platform. Kinda funny how cloud seems to be ramping up as an attack vector for Asha Sharma’s ridiculous intent to “entertain more than a billion people each day” when they already quietly shuttered the whole “everything is an Xbox” brand.

Meanwhile the second one isn’t any better frankly. Call me old fashioned, but I prefer to play the game as the developer’s intended. The whole thing reads like both an opportunity for useful user testing, but also a risk of aggressive of focus testing driven largely by ‘engagement metrics’. I can see the opportunity to do something interesting with it, but with a more cynical view it reads as another ominous and depressing future for video games.

Well, on that upbeat note, I’m out of here! Catch you next week.

AI and Games is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.