When Arc Raiders launched in October 2025, one of its biggest headaches was not the gameplay but the voices. The move follows ongoing criticism from players who were uncomfortable with the game's use of AI-generated speech when it released in 2025. Now, after becoming one of the most successful games Embark has made, reportedly passing 12.4 million copies sold in January 2026 and climbing to more than 14 million units by February, the studio is quietly correcting course.
Embark CEO Patrick Söderlund confirmed that "a lot" of the in-game voice lines in Arc Raiders are now recorded by human actors and there are fewer AI-generated lines in the game than there were at launch, noting that "we re-recorded some of the lines post-launch and made them with real voices." It's not a wholesale abandonment of synthetic audio, but it signals something important: there is a quality difference, and a real professional actor is better than AI.
This isn't Embark throwing in the towel on generative AI entirely. The studio's initial reasoning for using AI voices centred around development efficiency, because Arc Raiders is a live-service extraction shooter that receives regular updates, meaning recording new lines for every small addition could require bringing voice actors back into the studio repeatedly, and the AI system allowed the team to quickly generate placeholder or minor dialogue without scheduling new recording sessions every time the game changed. That's a legitimate concern for any live-service game trying to keep costs manageable while shipping regular content updates.
What's telling is how Söderlund frames the tool now. He says they look at AI "first and foremost as a production tool," allowing the team to "test things internally" and "test 15 different lines without recording them, and then we know what to record." In other words, it's a sandbox for deciding what's worth the expense of bringing a real actor into the booth. That's a far cry from using AI to replace voice talent altogether.
Some AI lines remain, though. For select situations, Embark pays actors for "approval to license their voices through text-to-speech" for audio lines that aren't as "essential to the immersion of the experience," mostly for location pings. These are background callouts, not main character conversations. It's a pragmatic distinction: a flat-sounding AI voice saying "enemy position north" is less jarring than an NPC you're supposed to care about delivering key dialogue with no emotion.
The counterargument deserves respect. Some observers have suggested that AI tools, when paired with proper human oversight and actor compensation, could actually expand opportunities for voice talent by enabling studios to produce more dynamic dialogue overall. But that argument assumes studios will consistently choose to compensate actors fairly and that financial pressure won't inevitably push toward minimising that cost. History suggests caution is warranted.
What makes Embark's move interesting is not that it contradicts some grand anti-AI principle. Rather, it's that the CEO is not trying to argue that AI and human performances are basically equal, instead openly acknowledging that trained actors deliver better results. That's a breach in the usual corporate tendency to defend every cost-cutting measure as equally effective as the traditional approach. The studio pays its actors for "all time spent" in the booth recording lines and continues to "bring many of them back as we carry on updating the game."
For a small team that beat the odds in a crowded multiplayer space, the shift suggests Embark recognises something the broader industry still debates: there are limits to where players will accept AI, especially when it touches the core experience they're paying for. Whether that boundary holds as AI tools improve, or as economic pressure mounts, remains an open question.