The impact of generative AI upon PC gaming has proven controversial, which is my balanced journalist way of saying it’s been horrible. Players are widely repulsed by genAI material, developers and even some publishers are increasingly wary of its temptations, and in a rush to build the requisite infrastructure, component shortages have ravaged the hardware market. Nonetheless, EVE Online devs Fenris Creations – formerly CCP Games – have become dead keen on robot brains, and what they might be might be able to think up for EVE itself.
Earlier this month, a newly independent Fenris announced a “research partnership” with Google DeepMind, the search giant’s AI research division, that would see DeepMind take a minority stake in the company while training its AI agents on a separate, offline version of the longstanding space MMO. Days later, Fenris CEO Hilmar Veigar Pétursson sat onstage with DeepMind co-founder Adrian Bolton at the annual EVE FanFest conference to discuss the partnership, in a presentation that left the concrete plans of what it means for EVE still broadly vague – yet seemingly against the run of wider sentiment, escaped any significant backlash from the game’s historically outspoken playerbase.
Much of the talk concerned DeepMind’s past work with games, including its Grandmaster-ranked StarCraft 2 bot and the use of 50 different points-based Atari games to effectively grade how a particular agent is learning. The simulated 3D movement in many games also, Bolton noted, has applications in self-driving car tech (though given the orbital maneuvering of EVE Online’s ship combat, I imagine this won’t be a focus here. Unless they want to teach driving around a really big roundabout). There were also some pre-emptive rebuttals to the notion, expressed by nobody in particular, that the training of machine-learned AI ‘players’ would kill off interest in multiplayer games, Pétursson citing chess’ lack of obsolescence after the Deep Blue supercomputer beat Garry Kasparov in 1997.
Lots about what EVE can do for DeepMind, then, but not much on what DeepMind can do for EVE. It wasn’t until an audience Q&A broke out that hints of game-specific applications began to emerge, the first of which was potentially using AI “coaching agents” to straighten out what players and devs alike call EVE’s “spaghetti code”. That’s the decades-old codebase that years of tweaking and staff turnover have twisted into dense, unknowable knots, complete with uniquely bizarre quirks. Many of the 1s and 0s governing EVE’s most fundamental functions are, for instance, still tied up in old code for Player-Owned Structures (POS), an abandoned space station mechanic that Fenris have been unsuccessfully trying to amputate for years.
“Coaching agents have been extremely helpful in identifying issues in our codebase,” Pétursson explained. “Because our codebase has now been written over 30 years, a lot of the context of the code is lost, because people have moved on or forgotten what they were doing at the time, and everything is always poorly documented… and coding agents are extremely helpful, in that we found a lot of memory issues, and we know increasingly they’re finding security issues. So, I would say yes, they, they are definitely assisting with repairing and replacing the speculative code. That is already happening. It’s mostly humans that actually do the replacement, but the discovery of where the problem areas are, that’s definitely already occurring.”
A more visible possibility would be building on the game’s existing proc-gen systems by adding genAI features to existing mission-giving NPCs – incidentally called Agents – in a change that Pétursson felt would make these characters “more alive.”
“The system that does most of the Agent missions was written 24 years ago, and there’s a lot sleeping there,” he told the FanFest crowd, adding that the original procedural generation tech was never particularly sophisticated despite being advanced for its time. “The fact that you are, in EVE, doing a mission for an Agent, it’s almost like the narrative itself fits very nicely. So all these elements of making EVE feel more alive, I think, are very exciting things to explore.”
For me, peering down at the talk from a darkened balcony like a joyless bat, it was hard not to recall the multiple reported cases of AI coding helpers deleting entire databases and/or knocking out Amazon Web Services. Or the conversation-cycling, uncomfortably dead-eyed spectacle of Nvidia’s experiments with genAI NPCs. If there was anyone else at FanFest who shared my worries, however, they weren’t making themselves known. Pétursson and Bolton left the stage to the same enthusiastic applause that met the Cradle of War expansion announcement, the Harpa concert hall’s seats slowly emptying from what looked close to a full house.
I don’t mention this as a criticism. I probably like EVE players more than I like EVE, as every fascinating story of industrial sabotage and economic warfare that comes out of this singular MMO is driven by the participants far more actively than the game itself. There’s also something endearing about the community’s long history of civil disobedience: the launch of expensive microtransations in 2011 (and Fenris/CCP internally publishing a stultifyingly tone-deaf magazine with the cover headline of “Greed is Good?”) sparked massive in-game protests, including the formation of vast spaceship conga line around a key trading system. In 2022, CCP’s attempt to sell a fully kitted ship, bundled with skill upgrades – breaking another monetisation taboo – prompted players and streamers to enact a ‘blackout’ campaign that, like a decade prior, ended with the offending items being taken off sale.
There were no signs of such discontent with the DeepMind deal at FanFest 2026. The question on whether AI could help fix the POS spaghetti code was submitted not as a joke, but because players genuinely want solutions to it. Far from a purely dev-side headache, pasta programming has been a longtime concern of players who feel it’s holding EVE Online back. And it’s not just those who ponied up for a ticket to Iceland – a similar comment on POS sits atop the EVE Online subreddit’s comment thread for the original Fenris/DeepMind announcement. Scroll further down, in fact, and there’s very little negative sentiment on AI integration at all, reactions varying between curious and disinterested but never veering into hostility.
FanFest attendees appeared happy to occupy the same spectrum. Mercfromabove, a fleet commander and shipbuilder for the powerful OnlyFleets alliance, was excited for the deal and its potential impact on EVE (though a voice recorder failure meant his exact words are lost to time, perhaps a cosmic punishment for me ambushing him with it). Dylana Aivo, 10-year EVE veteran who now runs training programmes for new players, felt Fenris and Google “maybe” hadn’t adequately explained what the partnership would entail for the game, but was content to watch it pan out.
“I was kind of surprised when I first heard about it,” she told me in a post-FanFest email. “I didn’t know much about DeepMind (and still don’t), but I think it would be interesting to see how they use it. This game has so much focus on involvement of real people with the economy, PVP content, etc. So I feel like there could be a lot of opportunity for AI while still keeping the spirit of the game.”
shutupandshave, a former CEO of mercenary megacorp Black Omega Security (though best known from his days as a musical propagandist for the infamous Goonswarm), is only slightly more anxious. “Let’s not let the AIs on, because they’ll absolutely destroy us!” he laughed. “But I think it’s really interesting. I think we’ve got some really, really unique data in EVE.”
“I will say my initial reaction was ‘Oh no, don’t train anybody on the stuff that goes on in EVE,’ y’know? This is like distilled Lord of the Flies, almost. And I don’t want any AI training on that! But CCP, traditionally, have partnered with quite serious players, and I can’t really think of any examples where it’s not made CCP better. So, go for it.”
shutupandshave – or ‘Uncle SUAS’ to fans of his tunes – recalled when Fenris/CCP teamed up with IBM to mend EVE Online’s then-shonky server infrastructure, as one such example of fruitfully making pals with Big Tech. EVE is also experimenting with a GPT-style tutorial chatbot, Aura Guidance, which is currently being trialled with a select few players. This technically uses AI, and predates the Google deal, though Fenris executive producer Snorri Árnason claims it doesn’t count as generative AI because it doesn’t generate anything – it only aggregates answers from existing tutorial materials. By all accounts, it’s been quietly welcomed.
Other, more recent adventures have been altogether less well-received; see the instantly wilted reaction to blockchain-entangled survival spinoff EVE Frontier when it was revealed in 2024. The DeepMind partnership, however, appears to have either delayed a backlash or dodged it entirely. For all the growing cynicism around AI use in games, from players and developers alike, EVE’s most dedicated players are at least willing to wait and see what happens.
According to Fenris community manager Peter ‘FC Swift’ Farrell, this doesn’t mean the game’s enthusiasts have lost their bite. “So EVE players… if I can speak freely, they have the biggest bullshit detectors on the planet,” Farrell told me, speaking freely. “Like, if you give them some sort of PR, saying-something-without-saying-something-speak, they will just immediately turn off to it.”
Farrell doesn’t share Dylana’s view that the implications of the partnership are unclear. “The Google DeepMind people have been really great at talking to the players about what this partnership means, about what the research entails,” he said. “It’s not like oh, it’s going to take over and recreate the art of EVE using some AI image generations or something. It was a research type of thing, and Google DeepMind have been doing stuff with games for a long time.
“They’ve got this huge, proven track record of approaching this in a very sane way, in a very respectful way of the product, and in how they approach things. I think when players saw this, they kind of looked into DeepMind a little bit more, and they could see that a lot of these guys have been doing stuff with games for so long, right? They did stuff with Starcraft, they’ve done stuff with Go, just their general approach to everything.”
Farrell added that Google have been proactive in responding to concerns, when they are expressed, over potential genAI overreach. “Players are curious about it, they want to know what this means, and there is probably some cautious scepticism in there as well,” he continued. “But so far, there’s no red flags that we’ve seen from them. And credit to the Google DeepMind people, as they’re watching this as well, because they want to make sure that they’re treating this with the respect and reverence that EVE Online has earned over the last 23 years. So if anything pops up, they’re quick to go out and be like ‘Yep, this is not what we’re doing. We’re doing this type of thing. Here’s how we’re doing it. Here’s what we’ve done with other games.’ And just pointing people in that direction.”
For now, it appears Fenris and DeepMind have done enough to convince EVE players to at least let them have a crack at making AI-powered changes. Which, to be fair, does involve the potential of addressing some of this community’s longest-lived complaints; if a coding agent really does manage to untangle that spaghetti, one can imagine even the most black-hearted of New Eden’s corporate murderers sending a nice hydrangea bouquet to Adrian Bolton’s house.
Still, it’s unlikely that the two companies inked a deal, allegedly worth millions, to clean up a codebase. And outside of a handful of examples, it remains unclear what a DeepMinded EVE Online will eventually look from the ground (er, space) level. Whatever’s in it for Fenris, it’s presumably worth the optics of joining with a company involved in very real warfare, its staff having recently begun unionisation proceedings in part to protest deals with the US and Israeli militaries.
Ironically, being able to work on a space MMO – even one with as much virtual backstabbing as EVE – might be just what dissatisfied DeepMind staff would like instead. Pretty much everyone else involved sounds willing, anyway.
