Google DeepMind has published a detailed account of its 15-year arc of AI research in games, culminating in its research partnership with Fenris Creations, the independent studio behind EVE Online, and laying out the staged research program the two organizations plan to run inside the EVE universe. The August 21, 2026 post connects the lab’s earliest reinforcement learning results to its current work on generalist agents and names the specific capabilities it expects the EVE environment to test.
The retrospective covers the chain of game results that built the lab’s research program: the Deep Q-Network that learned 49 Atari 2600 games from raw pixels, documented in a 2015 Nature paper; AlphaGo’s 2016 victory over world champion Lee Sedol; AlphaGo Zero and AlphaZero, which removed human data and generalized across chess, shogi, and Go; MuZero, which played without knowing the rules; and AlphaStar, which reached Grandmaster level in StarCraft II in 2019. The same foundations, the post notes, carried into AlphaFold, the protein structure system recognized with the 2024 Nobel Prize in Chemistry.
From Mastering Games to Operating Inside Them
The post frames a shift in the lab’s games research: away from systems that master a single game with a clear score, toward agents that can operate in any game world the way a person would. The vehicle for that is SIMA, the Scalable Instructable Multiworld Agent, which perceives what a player sees on screen and acts through ordinary keyboard and mouse controls with no access to game code.
SIMA 2, introduced in November 2025 with a Gemini model as its core, moved beyond the first version’s repertoire of over 600 language-following skills into reasoning about goals, conversing with users, and improving through self-directed play. Google DeepMind reports that SIMA 2 closes a significant portion of the gap to human task-completion performance across its training environments, including games it was never trained on, such as ASKA and a Minecraft research implementation. The lab’s own stated limits: the agent still struggles with very long-horizon tasks, works from a short memory constrained by the context window needed for low-latency interaction, and is available only as a limited research preview to a small cohort of academics and game developers.
What the EVE Partnership Adds
The Fenris Creations partnership, unveiled earlier this year when the studio behind EVE Online went independent, gives DeepMind access to a persistent world that has been running continuously since 2003. EVE Online’s single-shard universe holds thousands of players in one shared economy with real supply-and-demand dynamics and political structures built entirely by its players. Fenris Creations, the independent studio behind the EVE Universe, has spent more than two decades building one of gaming’s most persistent worlds.
The post identifies four capabilities the environment is expected to exercise: continual learning without forgetting, memory extending beyond current context windows, long-horizon planning over weeks or months, and complex multi-agent dynamics spanning cooperation, competition, negotiation, and economics.
“Together with Google DeepMind, we’re pushing into uncharted territory where AI must learn, adapt and remember on timescales that no other game environment demands,” Hilmar Pétursson, CEO of Fenris Creations, said in the post.
The partnership extends across three distinct environments. EVE Online offers the large-scale persistent universe; EVE Vanguard adds first-person, fast-paced tactical decision-making; EVE Frontier, with its programmable Smart Assemblies, presents a world where the rules themselves can change.
The research program follows a staged path: it begins in an offline instance of EVE Online separated from live players, moves through EVE Frontier to study how humans and agents coexist in a persistent world, and reaches the live games only when capabilities are mature. One system from the collaboration is already in front of players: Aura Guidance, launched in prototype on February 17, 2026, uses Gemini to answer new-player questions drawn from a vetted bank of real Rookie Help exchanges, and is running as a controlled A/B test measuring whether it improves new-player retention.
The longer arc, as the post frames it, runs from Atari’s raw pixels to a 23-year-old persistent universe: each environment chosen because it forces the next capability the lab believes a general agent needs. The EVE universe is the current test of that approach, and the offline sandbox is where the work now starts.
Credit: Source link
