Subnautica 2 Rejects Generative AI Despite Krafton's "AI-First" Mandate

The underwater survival sequel Subnautica 2 will launch without a single generative AI asset, despite publisher Krafton branding itself an "AI-first company" last October. Developer Unknown Worlds Entertainment made the position clear in interviews with Eurogamer, with creative media producer Scott MacDonald stating flatly: "No, we're not using generative AI in Subnautica 2 at all."

That's a notable stance in an industry where AI tooling has become nearly ubiquitous. Krafton's October announcement outlined a strategy to "prioritise AI as a central and primary means of problem-solving" across its portfolio. The company even offered voluntary resignation packages to staff uninterested in its AI-led future. Yet Unknown Worlds is opting out entirely.

MacDonald explained the distinction between traditional game AI and generative models. "When we mention AI, it's all traditional: a programmer has physically gone in and done it themselves, like [the] creature AI," he told reporters. The studio uses Unity Blackboard and hand-authored behavior trees for creature reactions and gameplay logic. (This is the kind of granular control that generative models simply can't replicate without introducing unpredictable bugs.)

Community manager Donya Abramo reinforced the position in a Discord post, per reporting from GamesRadar+. "We're not using generative AI to develop Subnautica 2," she wrote, pointing to established "classic AI" systems rather than prompt-based generation.

The technical reality matters here. Behavior trees require explicit programming of decision nodes, animation states, and environmental triggers. Every creature movement, every NPC interaction, every gameplay mechanic gets manually tuned. This creates predictable, testable systems. Generative AI, by contrast, introduces stochastic outputs that can break immersion or create unbalanced gameplay loops.

AI gameplay lead engineer Antonio Muñoz Gallego described the studio's approach as iterative, human-authored work. That's not just semantics. It means every asset has traceable authorship, every animation has intentional timing, every sound effect has deliberate placement. Players won't notice the difference immediately, but they'll feel it in the consistency of the experience.

Krafton's position appears to be permissive rather than mandatory. "Krafton has basically said 'here are all the tools. If you want to use them, you can'," MacDonald explained. "There's no mandate to use them or anything. We are free to say yes, or no, or whatever." The publisher made tools available but didn't force adoption.

Design lead Anthony Gallegos suggested Krafton understood the risk. "I think [Krafton] is smart enough to know that if we had to use those tools it could be disruptive to our process at this point. So a mandate would be harmful to the development of the game."

This isn't isolated. Project Windless, another Krafton title announced during February's Sony State of Play, also avoided generative AI for "narrative or content creation elements." The pattern suggests publisher-level tool availability doesn't automatically translate to studio-level adoption.

The legal and IP landscape around generative AI training data remains murky. Studios with mature pipelines face high integration costs when introducing generative workflows. Asset validation, QA investments, and pipeline modifications can outweigh short-term productivity gains. Unknown Worlds appears to have calculated the risk-reward ratio and decided against it.

Subnautica 2 launches in early access on May 14. The game's co-op features and expanded underwater world will test whether traditional development methods can compete with AI-accelerated production cycles. Early access reviews will reveal if the hand-crafted approach delivers the polish players expect.

Whether this positions Unknown Worlds as principled or inefficient depends on the final product. The industry will watch closely to see if other studios under publisher umbrellas follow similar patterns of selective adoption. Krafton's AI-first branding remains intact, but studio-level decisions tell the real story.

The question isn't whether generative AI works technically. It's whether it works for this particular project, this particular team, and this particular player base. Unknown Worlds has made its choice. The market will decide if it was the right one.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn