ARC Raiders robots aren't actually learning from your behavior, despite the popular theory

Since launch, ARC Raiders has had a theory circulating among players: that the game's robot enemies are adapting to human behavior, memorizing favorite hiding spots, and finding increasingly effective ways to hunt players down. Some even claimed the Arcs would soon start flushing "rats" out of their hiding holes, and players reported watching Leapers squeeze themselves into ever-tighter spaces.

The theory seems to trace back to a 2021 talk by Tom Soldberg, a machine learning software engineer at Embark Studios, in which he described how the studio used the technology to teach enemies to move and navigate the environment organically – rather than hand-animating every movement. The idea that the same tech might have been applied to the Arcs' hunting instincts sounded compelling and fit neatly into the game's lore, especially alongside videos of flying Arcs discovering new angles to kill players.

ARC Raiders design director Virgil Watkins shot down the theory, saying that any examples of seemingly intelligent enemy behavior are simply the result of hard work by Embark's AI designers and engineers:

"That's just us in the way we author them."

Watkins also clarified that Embark uses AI exclusively in the classic game development sense – nothing LLM-related.

"The machine learning is literally only for teaching them to walk and navigate the environment. It doesn't do any of their behaviours or their attacks or anything like that."

So if you've been noticing the Arcs getting smarter, that's the developers' doing, not evolving algorithms. The robots are frighteningly effective and their attack patterns continue to grow more varied – but they're not learning from your mistakes. At least not yet.