A decade after Pokémon Go became a global phenomenon, new revelations suggest that player-generated data has played a significant role in training advanced artificial intelligence systems—some of which may extend into military applications.
According to reporting from Ars Technica billions of real-world images captured by players were used by Niantic and its spinout Niantic Spatial to build powerful navigation models designed for robots, autonomous systems, and potentially drones operating in complex environments.
From Mobile Game to Massive Geospatial Dataset
At its peak, Pokémon Go encouraged millions of players to explore their surroundings, scanning real-world locations such as landmarks, parks, and public structures. These scans—often short video captures—were part of optional in-game features designed to enhance augmented reality experiences. However, those same scans later became valuable training data.
“Ground scans were one component to help train Niantic Spatial’s real-world foundation models,”
a company spokesperson told Ars Technica.
“The models are the product of that training, not a copy of or a means of accessing the underlying scans.”
The company reportedly trained its systems on roughly 30 billion images, many tagged with precise geolocation and orientation data. This allowed the creation of a “large geospatial model”—a detailed, AI-driven understanding of real-world environments.
The technology developed from this dataset is now being used in real-world applications. In early 2026, Niantic Spatial partnered with Coco Robotics to improve navigation for autonomous delivery robots operating in urban settings.
At the same time, a separate collaboration with Vantor has raised broader questions. Vantor, which has ties to U.S. defense and intelligence agencies, is working on systems capable of navigating environments where GPS signals are unreliable or deliberately blocked.
During the Defence Geospatial Intelligence conference in London, Niantic Spatial revealed that early testing of its integrated positioning system reduced navigation errors by up to 70 percent, achieving accuracy within 1.5 meters in some scenarios.
Ethical Questions Around Player Consent
While the technology itself is not inherently controversial, its origins have sparked debate. Many players were unaware that their gameplay contributions could indirectly support such developments.
Jeroen van den Hoven, a professor at Delft University of Technology, told Dutch publication Trouw,
“Without the large number of scans from all those gamers, the development of this system would never have progressed so quickly.”
He added that players
“have indirectly… made a contribution to military applications.”
This has led to concerns about informed consent. Critics argue that agreeing to a game’s terms of service does not equate to consenting to downstream uses of data in defense-related technologies.
As DroneXL editor Haye Kesteloo put it: “Consent obtained for a game is not consent for a weapons program.”
Company Response and Data Separation
Both Niantic Spatial and Vantor have emphasized that there is no direct sharing of Pokémon Go player data between companies. Vantor stated it does not have access to the original dataset, while Niantic Spatial clarified that its AI models are trained outputs—not raw data repositories.
Additionally, since May 2025, Pokémon Go has been owned by Scopely, meaning Niantic no longer collects new player data from the game.
Still, the historical use of player-generated scans continues to raise broader questions about how user data can be repurposed long after it is collected.
