Poffins for Predators: How Play Built the GPS-Denied Drone War
In late 2020, Niantic introduced a feature to its augmented-reality phenomenon Pokémon Go: players were encouraged to slow down at virtual landmarks (called PokéStops), record a 15-second video of their surroundings, and upload it to Niantic’s servers in exchange for trivial in-game rewards like virtual “Poffins” or extra Pokéballs. Over the next five years, players completed more than 30 billion of these ground-level, multi-angle scans. Now, an investigation by the Dutch newspaper Trouw has unmasked where those 30 billion sweeps of real-world physical terrain actually ended up. They have become the cornerstone of a partnership between Niantic Spatial (the company’s newly rebranded enterprise arm) and Vantor, a major US military contractor specializing in “unified spatial intelligence.” The result is a system designed to guide military drones, autonomous vehicles, and loitering munitions through electronic-warfare environments where traditional GPS is jammed or blacked out.
The underlying technology at play here is a Visual Positioning System (VPS). Unlike GPS, which relies on fragile satellite signals prone to spoofing and electronic countermeasures, a VPS relies on “matching” real-time video feeds with an existing 3D spatial foundation model. When a drone flies over an urban street corner or a ground robot navigates a narrow alleyway, it matches static obstacles—such as building facades, signs, fire hydrants, and architectural structures—against a spatial map. While satellite imagery can build crude elevation maps from above, it is famously blind to the ground-level vertical details that autonomous systems need to move. That gap can only be filled by ground-level photogrammetry. Historically, gathering this kind of data was an impossible logistical hurdle for the military, as deploying scan crews to walk public streets globally would trigger massive diplomatic and operational backlashes. Gamification solved this constraint overnight, turning millions of teenagers and unsuspecting civilians into an unpaid, global mapping corps.
Both Niantic Spatial and Vantor have responded to the public backlash with carefully worded, highly technical denials. They insist that raw player data or video files have not been directly transferred or sold to the military or Vantor under the agreement. This defense, however, is a semantic shell game. While the raw video files are not being copied into a military database, those 30 billion scans are the precise fuel that trained Niantic Spatial’s “Large Geospatial Model” (LGM). Under the joint partnership, Vantor is integrating its aerial-focused “Raptor” flight software with Niantic’s ground-based VPS to create a unified, self-healing coordinate grid. Once spatial features are baked into the deep learning weights of a foundation model, the raw footage is no longer needed. As privacy advocate and researcher Jelle van den Hoven pointed out to journalists, tracing spatial weights back to their origin is virtually impossible, allowing companies to issue technically true but fundamentally deceptive denials that are totally unfalsifiable.
The response from the developer and technology community has been a mixture of dark humor and profound exhaustion. In a widely shared Hacker News discussion, community members noted that this represents the ultimate realization of dual-use technology masquerading as toys. One user, @FrustratedMonky, joked grimly about “Pokémon Go Driven Drones… trying to catch every human and Transfer them to the Professor for Candy.” Others quickly debunked the corporate-backed notion that urban playground data is useless for military operations. While some industry professionals argued that gaming data doesn’t overlap with current, active desert-based theaters of war, commenters pointed out that this view is incredibly short-sighted. “And now there are detailed locations of our cities,” noted @drfloyd51. “We might not get killbots today, but we will get pacificationbots.”
This transition represents a terrifying erosion of the concept of “informed consent.” The millions of players who mapped their neighborhoods thought they were helping build a harmless collective playground for digital monsters. They had no way of knowing their hobby was a crowdsourced reconnaissance mission for the visual guidance systems of future autonomous weapons. In the era of the Large Geospatial Model, there is no longer any such thing as “harmless” consumer-facing spatial data. Every photo, every LiDAR scan, and every video snippet uploaded to a cloud server is being ingested, processed, and commercialized as the infrastructure for state-level surveillance and weapon systems. It is time we realize that the most dangerous Trojan horses do not look like military hardware—they look like free apps on our phones.
Sources
- Niantic Spatial: Pokemon Go scans aren’t used for drone training
- Pokémon Go data ‘exploited to develop navigation’ for military drones
- Niantic Spatial and Vantor Partner to Deliver Unified Air-to-Ground Positioning in GPS-Denied Areas
- Hacker News: Pokémon Go Scans Trained the Navigation Tech for Military Drones