Between 2021 and 2025, millions of Pokemon Go players walked their neighborhoods holding up their phones to scan parks, storefronts, statues, and street corners in exchange for rare in-game items. Those scans, 30 billion of them in total, did not disappear into Niantic's servers. According to an investigative report published June 6, 2026 by Dutch newspaper Trouw, they became the foundational training data for a Visual Positioning System that is now being licensed to a U.S. defense contractor for GPS-denied military drone navigation. For broader Pokemon coverage, see the Pokemon hub. For all gaming news, see the video games hub.
1. The Three Companies | Niantic Spatial, Vantor, and What Each Did
The Trouw investigation traces a data pipeline that passed through three distinct corporate stages. Understanding what each entity controlled is necessary to follow the chain of accountability.
DATA ORIGIN
Niantic Spatial
The engineering and data architecture entity that retained Niantic's mapping infrastructure after the Pokemon Go gaming division was sold to Savvy Games Group. Niantic Spatial owns the Visual Positioning System trained on 30 billion player scans and holds the licensing rights to that model.
DEFENSE RECIPIENT
Vantor (formerly Maxar Intelligence)
A U.S. defense and intelligence contractor holding major government contracts for tactical software and navigation systems for autonomous military assets. Vantor partnered with Niantic Spatial in December 2025 to integrate the VPS technology into its drone and autonomous vehicle navigation programs.
INVESTIGATION SOURCE
Trouw
The Dutch national newspaper whose investigative journalism team untangled the licensing agreements, corporate restructuring, and defense contracts connecting Pokemon Go's AR infrastructure to Vantor's military navigation program. Published June 6, 2026.
2. How Pokemon Go Became a Mapping Operation | AR Tasks and the VPS
Beginning around 2020 and 2021, Niantic introduced two features into Pokemon Go that were presented to players as engagement mechanics: AR Mapping Tasks and Powered-Up PokéStops. Both features required players to hold up their smartphones and record short multi-angle videos of specific real-world locations. The in-game reward for completing a task was rare items, virtual badges, or temporary PokéStop enhancements that improved loot drops in the immediate area.
The design of these tasks was optimized for coverage. Players were directed to walk around a location, rotate to capture multiple angles, and record detail at close and medium range. A person following the task instructions was, without knowing it, producing exactly the kind of overlapping multi-angle video data needed to reconstruct a precise 3D model of a physical space. Multiply that across millions of players in hundreds of countries over four years, and the result is a global 3D map of publicly accessible spaces built at essentially zero cost to Niantic.
That data was used to train a Visual Positioning System. Unlike GPS, which locates a device by triangulating satellite signals, a VPS locates a device by matching what its camera sees against a pre-existing 3D model of the world. Feed the camera a few recognizable features of an environment and the system can determine, within centimeters, where the device is positioned in physical space. It requires no satellite connection. It works indoors. It works underground. And it works in active electronic warfare environments where GPS signals are being deliberately jammed.
3. The Military Value | Navigation When GPS Is Dead
The defense community's interest in commercial VPS technology is a direct consequence of lessons learned in active conflict zones. In Ukraine, electronic warfare has neutralized GPS-guided weapons and drones at a significant rate. Military forces on both sides routinely deploy jamming and spoofing systems capable of cutting GPS signals across entire operational areas. A drone that navigates by GPS signal alone is vulnerable the moment it enters a jammed environment. Its guidance system goes blind.
A drone running an onboard VPS trained on Niantic Spatial's geographic database does not have this vulnerability. Using a thermal or night-vision camera, the drone looks at the terrain below it, matches what it sees against its internal 3D map, and determines its position independently of any external signal. The navigation is entirely self-contained. Electronic jamming cannot touch it.
"The players have indirectly, in a perhaps minimal yet effective way, contributed to military applications. Once data is absorbed into an AI neural network, it blurs into mathematical patterns, making it virtually impossible to untangle or trace back."
Van den Hoven's observation about the irreversibility of AI training is the core technical problem for anyone seeking accountability in this pipeline. Once 30 billion scans have been processed into a neural network's weights, those weights are mathematical abstractions. You cannot remove the Pokemon Go data from the model. It has ceased to exist as identifiable player scans and has become part of the model's generalized understanding of what the physical world looks like. There is no opt-out mechanism for data that has already been trained into a deployed system.
4. The Corporate Defense | EULAs, Loopholes, and the Baking Problem
Confronted by Trouw journalists, both Niantic Spatial and Vantor defended the data pipeline by pointing to the legal frameworks governing it. The defenses are technically accurate. They are also a case study in why technically accurate and ethically sound are not the same thing.
| The Corporate Defense | The Ethical and Practical Reality |
|---|---|
| Broad Consent Clauses: Niantic's terms of service grant a "transferable, sublicensable, perpetual license" to any uploaded AR content. | Lack of Informed Consent: Players consented to an expanded gaming feature, unaware that "sublicensable" included licensing the data to defense contractors. |
| Vantor's Denial: Vantor states it will not use live Pokemon Go user data in active defense software. | The Baking Loophole: Vantor declined to deny whether the foundational AI model it licensed from Niantic Spatial was previously trained on those 30 billion scans. The live data question is a different question from the training data question. |
| Clean Corporate Break: Niantic Spatial notes that after Scopely's acquisition of the gaming division, active player data is no longer shared with them. | The Data Already Exists: The 30 billion scans collected between 2021 and 2025 have already finished training the models. The pipeline is closed on the collection side. The output models are already deployed. |
The "baking loophole" is the most legally significant aspect of Vantor's response. The company was asked a specific question: did you use Pokemon Go player scans to train the navigation AI you are deploying for military applications? Vantor's answer addressed live data only. It did not address training data. That is a non-answer to the question that was asked, and Trouw's investigators noted the distinction explicitly.
5. Why This Matters | Consumer Apps as Surveillance Infrastructure
The Pokemon Go investigation is not an isolated incident. It is a documented example of a structural dynamic that exists across the consumer app landscape. When a game or application gamifies real-world data collection, it can function as a crowdsourcing operation that bypasses the regulatory frameworks designed to govern data collection for surveillance and defense purposes. Players are not research subjects and are not paid for their contributions. They are not informed that the data they generate has commercial value beyond the in-game reward they receive. The EULA is the only instrument that governs the entire transaction.
The regulatory frameworks that exist for defense contractor data collection, including requirements for informed consent, geographic restrictions on surveillance data, and oversight of data use for weapons development, do not apply when the data enters the pipeline through a consumer entertainment product. The gamification layer is, functionally, a regulatory bypass.
Whether any of this constitutes a violation of existing law depends on jurisdiction. European data protection regulators are likely to examine the Trouw findings given the GDPR's requirements around purpose limitation: data collected for one purpose cannot be freely repurposed for another without renewed consent. The Niantic Spatial response relies on the argument that the EULA constituted blanket consent. GDPR regulators have historically been skeptical of that argument when applied to sensitive secondary uses that users could not have reasonably anticipated at the time of consent.
For all Pokemon coverage on OzoneNews, see the Pokemon hub. For Nintendo Switch 2 news including Pokemon Pokopia, see the Switch 2 hub. For all gaming coverage, see the video games hub.
Reported by Jack Sterling, Technology and Gaming Reporter, OzoneNews. Source investigation: Trouw, June 6, 2026. Last updated June 12, 2026.
Sources & References
- [1] Trouw | Original Investigative Report on Pokemon Go Data and Military Technology
- [2] PC Gamer | Pokemon Go Data Was Used to Help Train AI Systems Being Developed for Military Drones
- [3] Game Developer | Report Indicates Pokemon Go Video Scans May Have Been Used for AI Drone Training
- [4] DroneXL | 30 Billion Pokemon Go Player Scans Used to Train Niantic Spatial's Drone Navigation System