Arc Raiders AI Voices: The Backlash, the Re-Recordings, and What Is Still Synthetic

Key Takeaways

• Arc Raiders launched on 30 October 2025 for PlayStation 5, Xbox Series X/S and PC via Steam and the Epic Games Store, priced at $39.99 — not in late October 2026. The November 2026 coverage is about the game’s first anniversary and a new behind-the-scenes video revisiting a year-old controversy.
• At launch the game used a hybrid: most dialogue was recorded by real actors, but some mission and briefing lines, and ping-system lines, were generated by an in-house text-to-speech model trained on the game’s own actors’ recordings.
• The voices were licensed from the actors under contract, and the actors were paid for booth time and separately for licensing approval. The objection was about quality and immersion, and about AI replacing actors — not about unlicensed voice cloning.
• On 13 March 2026, CEO Patrick Söderlund said the studio had already re-recorded some lines post-launch with real voices, adding: “A real professional actor is better than AI; that’s just how it is.”
• On 3 November 2026, Embark published a behind-the-scenes video, The Evolution of ARC Raiders, Season 2 Episode 5, “Sound & Music”, in which the studio said the criticism was fair and that it is phasing out synthetic lines with human recordings “over the coming years” — with no deadline given.
• AI voices have not been fully removed: the ping system still uses licensed AI-generated lines, and Embark has not published a list of which lines are recorded versus generated, so no count or share can be stated.
• Search interest in “arc raiders player backlash ai” showed a breakout rise over seven days, and “embark studios arc raiders ai voice” rose 4,350%, according to the trend data behind this story — rises in search interest, not measures of search volume.

First, the date: this is a one-year-old controversy

Much of the confusion around this story comes from the calendar, so it is worth settling it first. Arc Raiders launched on 30 October 2025, for PlayStation 5, Xbox Series X/S and PC via Steam and the Epic Games Store, at $39.99. The game is an extraction shooter from Embark Studios, and the AI-voice controversy began at launch — a full year before the coverage that brought it back into the news.

What happened in November 2026 is an anniversary revisit. On 3 November 2026, Embark published a behind-the-scenes video, The Evolution of ARC Raiders, Season 2 Episode 5, “Sound & Music”, hosted by Danny O’Dwyer, in which the studio returned to the voice issue and acknowledged the criticism. Coverage followed the same day from Eurogamer via GamingBolt, PC Gamer, Kotaku and Pixel Twelve. The story, in other words, is not that a new game launched with AI voices. It is that a year-old game, a year after its launch, finally talked openly about a decision it had stayed quiet about.

That distinction matters, because it changes the shape of the story. This is not a launch scandal. It is a slow reckoning: a studio that used AI voices, said little about it, fixed some of it quietly, and then, a year later, explained itself on camera.

What the hybrid voices actually were

At launch in October 2025, Arc Raiders used a hybrid approach. Most dialogue was recorded by real actors. But some mission and briefing lines, and lines in the ping system — the quick contextual callouts players trigger to mark items, enemies and locations — were generated by an in-house text-to-speech model trained on the game’s own actors’ recordings. The pings were a patchwork of recorded and generated lines, and players could not reliably tell which was which.

That last detail is important. The text-to-speech model was trained on the actors themselves. These were not generic synthetic voices, and they were not the voices of people who had never agreed to be recorded. The TTS voices were licensed from the actors under contract. The actors were paid for their time in the recording booth, and they were paid separately for licensing approval — for agreeing that their voices could be used through text-to-speech for certain lines.

The same approach had been used in Embark’s earlier shooter, The Finals. In October 2025, chief creative officer Stefan Strandberg said Arc Raiders’ AI-voice usage was “on the same level” as The Finals: “We use a combination of recordings of real actors and TTS.” The studio’s position, from the start, was that this was a production tool, not a replacement strategy — a way to cover lines that changed late in development or that existed in too many variants to record one by one.

That is a real production problem, and the studio’s explanation of it deserves a fair hearing. But it is also a problem that players experienced as a quality failure, which is where the backlash began.

Why players objected

Players and reviewers objected from launch, and the objection had two strands that are often blurred together. The first was about quality and immersion. The AI lines sounded lifeless and flat, according to players and reviewers, and they broke immersion. Eurogamer’s review flagged the issue at launch. In a game built around atmosphere — a ruined, rain-soaked world where every sound cues danger — a flat, synthetic voice is not a minor flaw. It is a crack in the thing the game is selling.

The second strand was about AI replacing actors. Players and journalists argued that using generated voices, even licensed ones, took work away from performers and set a precedent that other studios would follow. That concern was about the direction of the industry, not about this one game.

What the objection was not about is equally important. It was not about unlicensed voice cloning. The actors had consented, and they had been paid — for booth time and for licensing approval. There is no allegation in the coverage that Embark used anyone’s voice without permission, or that it trained its model on recordings it did not own. The distinction between licensed, consented AI voices and unlicensed cloning is the difference between a quality debate and a rights violation, and blurring it does a disservice to both sides. The actors were partners in the arrangement, not victims of it, and at least one of them has said so on camera, as the section below explains.

It is also worth noting that the quality problem cut both ways. Audio director Bence Pajor laughed in the November 2026 video that players complained about AI even on human-recorded lines — “you can’t even tell, so why does it matter?” That is a fair point from the studio’s side, and it is also a sign of how thoroughly trust had broken down. Once players believed the voices were synthetic, they heard synthetic voices everywhere, including where there were none.

The timeline: launch, a quiet fix, and a public reckoning

The story unfolded in three acts across a year.

Act one: 30 October 2025. Arc Raiders launches with hybrid recorded and AI-generated voices. Backlash from players and journalists begins immediately, focused on the flat quality of the generated lines and on the principle of AI replacing actors.

Act two: 13 March 2026. In an interview with GamesIndustry.biz, conducted by Alex Forbes-Calvin, CEO Patrick Söderlund says the studio had already re-recorded some lines post-launch with real voices. The interview was reported by Polygon, KitGuru and TheGamer, and Söderlund’s words were blunt. “We re-recorded some of the lines post-launch and made them with real voices. There is a quality difference. A real professional actor is better than AI; that’s just how it is.” He also explained the payment arrangement in detail: “We pay our actors for all time spent with us in the booth and continue to bring many of them back as we carry on updating the game. For select usage, we also pay them for the approval to license their voices through text-to-speech for lines that aren’t as essential to the immersion of the experience, mostly ping system audio.”

Note the word “some.” In March 2026, the studio said it had re-recorded some lines. It did not claim to have replaced all of them. That careful wording matters, because later reporting would be less careful.

Act three: 3 November 2026. Embark publishes The Evolution of ARC Raiders, Season 2 Episode 5, “Sound & Music”, hosted by Danny O’Dwyer. The video says that after release there was an outcry from players and journalists, and that the studio stayed silent and went back to work re-recording lines with the actors. One team member says they re-recorded all the mission lines with actors “as far as possible.” The ping system, however, remains partly synthetic.

That phrase — “as far as possible” — is doing a lot of work. It is not the same as “all.” One outlet has claimed that all quest lines were re-recorded, but it notes that the extent is unverified, and the March 2026 reporting says “some” while the November 2026 video says mission lines “as far as possible.” The honest summary is that some launch lines, including mission lines as far as possible, have been re-recorded with real actors. Anything stronger than that goes beyond what the sources establish.

What the November video actually said

The video is striking less for what it reveals about the technology than for its tone. Embark does not defend the launch decision. It concedes the criticism.

Executive producer Aleksander Grøndal says, in the video: “Every time you’re testing out a new technology, you don’t really know how people will react to it… I think there was some really fair criticism in there. There’s no point in trying to be defensive; it’s just what’s best for the game at the end of the day.”

Chief creative officer Stefan Strandberg goes further, addressing the replacement fear directly: “We are not in the industry of replacing actors. Actually, I think quite the opposite. We should celebrate the performance and give them more space.” The studio says it is phasing out synthetic lines with human recordings “over the coming years” — but no deadline was given, and no policy banning AI voices was announced. The studio still defends text-to-speech as a production and prototyping tool.

Audio director Bence Pajor explains why the ping system stays AI. The list of pingable items and features keeps growing, and recording every variant is impractical. His explanation, quoted by GamingBolt from the video, is worth reading in full: “[The actors] have been here all along. But we haven’t always used them to record the actual thing you’re going [to hear] because that’s only been solidified super late. We made a lot of training data from their voices so we could prototype as well with their voice.” And on the ping system specifically: “Every new item that appears in the game, every new feature, everything needs to have a lot of dialogue because you can ping it… To record the training data for the ping system becomes too impractical to record every possible version of something. I mean, the actors would just go crazy if they had to read all of that. That’s not feasible because the game has been changing so fast and we can’t call them in every time we just change a feature.”

That is a practical argument, and it has a practical logic. A live game changes constantly. New items, new features and new map elements each need dialogue, and each variant multiplies the recording burden. Calling actors back into the booth every time a feature changes is slow and expensive. A licensed text-to-speech model, trained on those same actors, can generate the new lines immediately. The question is whether that convenience is worth the quality cost — and the studio’s own CEO has now said, on the record, that the quality cost is real.

The studio’s broader position on generative AI has been consistent. Design director Virgil Watkins said earlier in 2026 that the team has “no plans to ‘open the floodgates'” on generative AI, and Embark says it does not use generative AI “in other domains across the game”, according to Strandberg in October 2025. The voice work was presented as a contained exception, not the start of a wider programme.

The actors’ side of the arrangement

The video also gives the actors a voice in the debate about their own voices, and what they say complicates the simple version of the story.

Voice actor Diana Gardner, who plays Celeste, says the arrangement felt “freeing” for scheduling, and that the contract “seemed so fair” it was “a no-brainer.” That is not the testimony of someone who feels exploited. It is the testimony of someone who was paid, consulted and brought back repeatedly as the game was updated.

But Gardner also raises a question that has no easy answer. On ownership, she asks: “Can I take ownership for the performance? … This is hours of work that I’ve put in, but can I take ownership for the final product? No.” That is the heart of the licensing dilemma. An actor can be paid fairly, consent freely, and still feel that something essential has been separated from them — the performance exists, it is built from their hours of work, and yet the final product is not quite theirs. It is generated, not performed, and the actor’s name sits at a distance from the thing players actually hear.

Her fellow actor Jo Rideout plays Shani, and Celeste’s in-game appearance was also updated to resemble Diana Gardner in October 2026, according to Voice Over Herald — a small detail that suggests the studio has been working, visibly, to put the human performer back at the centre of the character.

The actors’ perspective is a useful corrective to the loudest versions of this debate. The issue was never that Embark stole voices. The issue was that even a fair, licensed, consented arrangement can produce a worse experience for players and a stranger relationship between performer and performance. Consent solves the legal question. It does not solve the artistic one.

What has changed — and what has not

A year after launch, the honest status report is mixed, and it should be stated plainly.

What has changed: some launch lines have been re-recorded with real actors, including mission lines “as far as possible” according to the November 2026 video. The studio has acknowledged the criticism publicly and on camera, in language that concedes the point rather than disputing it. The chief creative officer has said the studio is not in the business of replacing actors, and the studio says it is phasing out synthetic lines with human recordings over the coming years.

What has not changed: AI voices have not been fully removed. The ping system still uses licensed AI-generated lines, because the studio says recording every variant is impractical. Embark has not published a list of which lines are recorded versus generated, so no count or share of synthetic lines can be stated. No policy banning AI voices has been announced, and no deadline has been given for the phase-out. The studio still defends text-to-speech as a production and prototyping tool.

That is a partial fix, honestly described. It is not a reversal, and it is not a whitewash. It is a studio saying, in effect: we were wrong about quality, we are fixing what we can, and we are keeping the tool where we think the alternative is unworkable. Whether that is enough is a judgement readers can make for themselves — but it should be made on the basis of what actually happened, not on the basis of the strongest claim in either direction.

The sceptical view

Not everyone was convinced by the November video. Kotaku’s John Walker was sceptical, framing the video as rationalisation — a studio explaining away a decision it had already made, rather than reckoning with it. That scepticism is worth reporting, because it captures a real tension in the story. A behind-the-scenes video is a piece of studio communication, not an independent investigation. It is produced by the company whose decision is under discussion, hosted by a presenter the company invited, and edited by the company itself. The concessions in it are real — the quotes from Grøndal and Strandberg are on the record — but the framing is the studio’s own.

The sceptical reading also has a practical point. “Phasing out synthetic lines over the coming years” is a promise without a deadline, and promises without deadlines are easy to keep and easy to break. If the ping system stays synthetic indefinitely because the game keeps changing, then the phase-out may amount, in practice, to the mission lines only — which is a real improvement, but not the full removal that some players asked for.

On the other hand, the studio’s actions before the video lend the words some weight. The re-recordings began before March 2026, months before the anniversary video, and they were disclosed in an interview rather than announced as a marketing beat. A studio that intended only to rationalise would have had less reason to start fixing the lines quietly, and more reason to wait for the video. The record shows work first and explanation second, which is the better order.

What this means for AI voices in games

Arc Raiders is a useful case study because it sits in the middle of the debate, not at either extreme. It is not a story about unlicensed cloning, and it is not a story about a studio refusing to listen. It is a story about a licensed, consented, paid arrangement that still produced a bad experience — and about a studio that took a year to say so publicly.

Three lessons stand out. First, consent is necessary but not sufficient. The actors agreed, were paid, and were brought back. Players still heard the difference, and they still objected. A licence does not make a flat line sound alive.

Second, the production argument has limits. The ping system is a genuine scaling problem, and Pajor’s explanation of it is credible. But the same logic, applied more broadly, would justify synthetic voices everywhere a game changes fast — which is everywhere. The line between “impractical to record” and “inconvenient to record” is a judgement call, and it is one that players, not just studios, will want a say in.

Third, transparency matters more than the technology itself. Embark stayed silent after the launch outcry and went back to work. The work was real, but the silence let the story harden into something worse than the facts: a studio secretly replacing actors with AI. A year later, the video corrected the record, but a year is a long time for a misunderstanding to live. The wider games industry is now writing AI policies — the UK Games Expo, for example, has introduced new AI rules for exhibitors in 2027, covered in our report on the UK Games Expo AI Policy: What the New Rules Mean for Exhibitors in 2027 — and disclosure is likely to become a bigger part of those rules, not a smaller one.

For players, the practical takeaway is narrower. If you play Arc Raiders today, you will hear more real actors than you did at launch, and the mission lines have been re-recorded as far as possible. You will still hear synthetic voices in the ping system. The game does not tell you which is which, and the studio has not published a list. That is the current state of play, a year after launch, and it is the state of play the anniversary video was made to explain.

The same questions are spreading beyond games. Voice synthesis is improving fast, and the tools for choosing and evaluating AI-generated voices are proliferating — our guide on How to Choose an AI Talking Photo Generator: Features, Limitations, and Use Cases covers how to judge synthetic voice tools on quality and limits. The Arc Raiders story is a reminder that the judgement cannot stop at the tool. It has to include the contract, the credit, the quality, and the honesty about which lines are which.

The bottom line

Arc Raiders launched on 30 October 2025 with a hybrid voice system: real actors for most lines, licensed AI-generated voices for some mission and briefing lines and for the ping system. Players and reviewers objected from the start — not because the voices were unlicensed, but because they sounded flat and broke immersion, and because they feared AI replacing actors. The studio re-recorded some lines quietly, said so in March 2026, and then, in a behind-the-scenes video on 3 November 2026, admitted the criticism was fair and promised to phase out synthetic lines over the coming years.

That is a real concession, and it is also an incomplete one. The ping system is still synthetic, no deadline has been given, and no list of recorded versus generated lines has been published. The story of Arc Raiders’ AI voices is, in the end, a story about the gap between a fair contract and a good performance — and about how long it can take a studio to close that gap in public.

Sources: Polygon, Arc Raiders has fewer AI voices now; GamingBolt, Arc Raiders studio acknowledges really fair criticism about its use of generative AI; Pixel Twelve, Arc Raiders AI voice pings, Embark; Wikipedia, ARC Raiders.