Google DeepMind announced a research partnership with Fenris Creations, the independent studio behind the EVE Universe, to develop AI agents capable of learning and adapting inside persistent, player-driven game worlds. The collaboration extends across EVE Online, a massively multiplayer space simulation running continuously since 2003, and EVE Vanguard, a first-person tactical shooter set in the same universe.
The partnership marks a shift from DeepMind's earlier work mastering games with clear objectives — such as Atari, Go, and StarCraft II — toward building agents that understand open-ended environments without predefined scores or rulebooks. The lab's Scalable Instructable Multiworld Agent (SIMA) 2, powered by Gemini models, already demonstrates human-like play across 3D titles including No Man's Sky, Valheim, and Hydroneer using only on-screen pixels and natural language instructions.
What's new
- DeepMind partners with Fenris Creations across the EVE Universe (EVE Online and EVE Vanguard).
- Focus on continual learning, long-horizon planning, memory across extended timescales, and complex multi-agent dynamics.
- SIMA 2 acts as an interactive companion using ordinary keyboard and mouse inputs — no game APIs or source code required.
- Genie 3, a general-purpose world model announced separately, generates interactive environments from text descriptions.
EVE Online's single-shard universe, with its player-driven economy spanning thousands of star systems and a history shaped by alliances, conflicts, and diplomacy, presents a uniquely demanding testbed. The environment requires agents to acquire new skills without forgetting previous ones, reason over weeks or months, and navigate emergent social behavior at scale — capabilities DeepMind believes are essential for frontier AI systems that learn continuously from experience.
"EVE Online was envisioned from day one as a sandbox of lasting consequences, shaped by its players," said Hilmar Pétursson, CEO of Fenris Creations. "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, while helping us understand how humans and AI can coexist in a virtual environment before we have to contend with the same questions in real life."
Why it matters
If successful, the research could produce AI companions and non-player characters that genuinely understand game worlds and adapt in real time without re-scripting — transforming both gameplay experiences and game development workflows such as QA testing. The same capabilities — continual learning, long-horizon reasoning, and multi-agent coordination — are directly relevant to real-world applications where agents must operate in open, evolving environments.
Our take
DeepMind is betting that the messy, player-driven complexity of EVE Online will force breakthroughs in memory and continual learning that sterile benchmarks cannot. The partnership's value hinges on whether agents can truly persist and adapt across months of gameplay — not just perform well in short evaluation windows.
