Arc Raiders’ use of AI highlights the tension and confusion over where machine learning ends and generative AI begins
Like Embark Studios’ 2023 shooter The Finals, its new game Arc Raiders, out today, comes bearing an “AI generated content disclosure” on its Steam page. “During the development process, we may use procedural- and AI-based tools to assist with content creation. In all such cases, the final product reflects the creativity and expression of our own development team,” reads the disclosure.
Despite multiple interviews with key Embark personnel in recent days, it’s hard to pin down exactly what pieces of Arc Raiders result from what sort of AI.
Watch On
Much of the unease around AI is also simply about it putting people out of their jobs
If your definition of generative AI is exclusively using tools like ChatGPT—which are built by ignoring copyright to scrape vast swathes of the internet for “training data” and then arguing they should be allowed to do so because they couldn’t exist otherwise—then it seems that Arc Raiders is in the clear. And those tools do represent the most dramatic version of what AI critics find reprehensible about the technology: that what it “creates” is simply an amalgamation of stolen, remixed and uncredited human art.
But much of the unease around AI is also simply about it putting people out of their jobs—jobs where their own creative input could’ve both kept them fed and made the end product better. Embark Studios may have saved money and time by not “needing to have [a voice actor] come in every time [they] create a new item for the game,” but I suspect most voice actors would be happy to get paid to say a bunch of item names every month or two. That repeat gig is now gone.
Isn’t using actors’ voices as training data to create audio barks and pings that they never said, by definition, generative? The studio may have found actors willing to sign contracts allowing that use, but if anything, that proves AI critics’ fears are well-founded.
Videogame voice actors in the SAG-AFTRA union went on strike for a full year, from July 2024 to July 2025, in part to secure a contract that limited how their voices could be used in conjunction with AI. The union won: the publishers they were fighting, including Activision, EA, and Take Two agreed to remove a “one-time payment option” for digital replicas of actors’ voices, Variety explained. In the final agreement, “performers who consent to the use of a digital replica receive compensation for that use comparable to what they would have received had they directly provided services.”
As far as I can tell, Embark Studios—which is based in Sweden—did not work with actors from the US-based union on Arc Raiders.
The studio’s interest in AI tech doesn’t end with text-to-speech, but every interview that touches on the subject seems to make it that much harder to pin down. Speaking to Edge magazine, Embark founder Patrick Söderlund said that they’d developed AI tools that let them create a 3D model of a game from a YouTube video; executive producer Aleksander Grøndal then told Rock Paper Shotgun that that was “a research project… not something that we’re using in the game now.”
Söderlund has also spoken about the company’s machine learning efforts to help animate its AI robots. Which isn’t “generative” AI—but it still reflects the core tension of AI tools, which promise to save human workers time and tedious effort up until the point that they can (oops!) maybe replace them altogether.
“The foundation of Embark was, if we’re going to try to build a studio, we can’t compete with the likes of EA and Activision and other large companies because we don’t have the same financial means to do that. We then realized that we had to get to a place where… we have to go back and take a look at how we develop a videogame,” Söderlund said on the podcast The Game Business this week.
That’s when we came into procedurally generated content, using AI and machine learning to some extent, even in the content creation pipelines
Patrick Söderlund
Söderlund came to the conclusion that for Embark’s 300 person team to compete with those triple-A studios and “Asian live game developers who churn out content,” they’d have to aim to be “100 times faster” than they were used to:
“If our goal is to do it 10 times faster, we’re going to try to fix conventional development methods and that may lead to us being twice as fast if we do a good enough job. If you ask for 100 times faster, you have to take what you know and throw that away and start really thinking about what type of modification and transformation we need to go through to completely change the way we approach this.
“That’s when we came into procedurally generated content, using AI and machine learning to some extent, even in the content creation pipelines, to what types of tools we build. A lot of these tools we’ve been working with for so long stem from software that’s 25-30 years old. We may have to write our own programs to get to where we want to.”
Söderlund added that it would be “arrogant” to say that Embark today is 100 times faster than any other studio, but does believe they’ve “created a way to build products and games and update them that [he doesn’t] think many others can compete with.”
At a time when mass layoffs in the game industry are rampant, there’s clear appeal to building games at the fidelity players have come to expect with a smaller team than it takes to create an Assassin’s Creed or Call of Duty. But it’s also hard to square the Embark team’s reassurances that they don’t use generative AI—and don’t want to remove humans from the equation—with that goal of being 100 times faster. And from there, to connect the dots to how the creative process might suffer from fewer human animators or voice actors bringing their own perspectives and talents into the fold. At what point on that slippery slope do the tools that aid efficiency instead begin to cause erosion?
I’m not yet equipped to answer that question, and it doesn’t seem like the game developers praising AI—generative or not—are either.
