Your game is a training world. AI labs will pay for it.
AI labs need realistic worlds to train their agents. Your finished game is one. benned makes it agent-ready, hosts a private copy for each lab and runs it. You sell under your own name and earn on every agent that plays.
AI labs pay for private, managed copies of realistic games to train their agents in. A studio licenses a finished game to one lab for a defined use; benned makes the game agent-ready, hosts that lab's copy on dedicated hardware and operates it. The lab pays an integration fee, a monthly service fee and a fee per concurrent agent, and the studio earns an agreed share of all three — without funding the engineering, releasing anything publicly, or touching its live game.
AI labs are buying worlds, not just data.
AI used to answer. The current generation has to act — operate software, drive a vehicle, run a project — and acting is a different skill, because the consequence of a decision arrives long after the decision. You cannot read your way to it. You get it by trying, failing and trying again, millions of times.
The real world is too slow, too expensive and too dangerous to practise in, so labs need a world that behaves like it, resets in a second and keeps score. Building one takes years. Games already are those worlds, so labs have started paying for private, managed copies of them. Google DeepMind, for example, took a stake in the studio behind EVE Online in 2026 and runs an offline version of the game on its own servers for research.
Anthropic leaders discussed spending over $1 billion on training environments in a single year.
Source: TechCrunchGoogle DeepMind took a minority stake in the studio behind EVE Online and runs the game offline for research.
Source: InvestGameFrontier labs, robotics companies and world-model builders. Few buyers, large budgets.
Source: Wing VCYou keep the customer. We do the work.
Labs are wary of middlemen that bundle many games and see what every lab does. So benned stays in the background. The lab contracts with you; benned is your technical partner.
- 1
Fit check
We assess whether your game suits AI training: realism, depth of choices, how customisable it is. Free, under NDA.
- 2
Agent-ready build
We make your game run without a screen, faster than real time, with resets and scoring. With your engine access, or as a black box on top of the normal build.
- 3
You sell, we host
Each lab gets its own locked copy on dedicated hardware in the Netherlands. We meter usage and handle uptime and support.
- 4
Recurring revenue
Labs pay an integration fee, a monthly service fee and a fee per concurrent agent. You earn on all of it.
One game, three labs, one studio getting paid.
What changes hands, who does the work, and where the money comes from. Illustration, not a screenshot.
A simulator the studio already built and already shipped. Its commercial life may be over; its systems are not. Nothing about it changes from here on.
The best bots in the world, playing your game.
A lab's agents are non-human players by construction, running at scale inside your systems and looking for the cheapest path through your rules. That is the exact signal an anti-cheat system has to catch, and it is a stress test no QA team can reproduce.
None of it reaches you automatically, and we would rather say so than imply otherwise: benned stores no agent data and the lab holds its own. If you want behavioural findings back, it is a term you negotiate into the licence and the lab has to agree to it. Ask early — it is far easier to put in before a deal than after.
Further out, the same engineering points the other way. A studio that has made one game legible to agents has learned something it can build with, for players as much as for labs.
Where that clause would sit- Exploits found at a scale you cannot buyThousands of instances probing your systems for the shortest route, continuously, for as long as the lab trains.
- Non-human play, labelled by constructionYou know every one of those sessions was an agent. That is the hard part of a detection dataset, and it comes for free.
- Only if you negotiate for itIsolation cuts both ways: the same rule that stops us watching a lab stops us handing you its runs. This is a licence term, not a feature.
Which games labs want most
The more realistic the game, and the more situations and actions it offers, the more labs want it. Release date matters less than depth.
- Driving, flight and transport simulators
- Construction, farming, logistics and city builders
- Management and economic simulations
- Open worlds with physics and many ways to reach a goal
- Older titles with a deep simulation, even if sales have faded
Very little. Your build never leaves controlled hardware.
No public release, no new players, no change to your live game. Every lab environment is physically separated, and agent data stays with the lab.
- Your name, your contractYou license to the lab directly. benned is your processor and partner.
- No source code requiredEngine access makes it faster and richer, but a licence and a build are enough to start.
- Upfront work paid by the labThe integration fee covers our engineering, so you invest nothing.
Before you get in touch
Can I license my game to AI labs without sharing source code?
Yes. benned can run your normal build as a black box: the agent sees the screen and sends keyboard and mouse input, like a player. Engine access lets us add faster-than-real-time play and richer scoring, but it is optional.
Who signs the contract with the AI lab?
You do. The lab licenses your game from you. benned is your technical partner that builds, hosts and operates the environment, so the customer relationship stays yours.
How do studios earn money from this?
Labs typically pay an upfront fee for the engineering, a monthly managed-service fee and a fee per concurrent agent. You receive a share of all three. We agree the split per game.
Will this affect my players or my live game?
No. The training copies run on separate, dedicated hardware with no connection to your live servers or player base.
Does a studio get anything besides money?
Potentially, and it is worth asking for. A lab’s agents are non-human players at scale, probing your systems for the cheapest path — which is exactly what anti-cheat and bot detection need to learn from. It is not automatic: benned stores no agent data, so access to behavioural findings is a term you negotiate into the licence and the lab has to agree to it.
Does an old game still have value?
Often, yes. Labs care about realism and depth of the simulation more than release date, so a back-catalogue simulator can earn again.
Start with one game and one pilot.
Studios: tell us which titles you could license. Labs: tell us what your agents need to learn. We reply within two working days, and nothing you send goes further than us.