An artificial intelligence model trained on data harvested from millions of Pokémon Go players is being used to help military drones navigate war zones without relying on satellite signals, raising urgent questions about how civilian data is repurposed for defence applications.
Pokémon Go, the augmented reality game that took the world by storm in 2016, allowed players to chase and capture virtual creatures on their phone screens while moving through real-world streets, parks and buildings. By 2018, the game had been downloaded more than 800 million times worldwide. A 2021 update introduced so-called “AR Mapping tasks” and “Powered-Up PokeStops” that rewarded players with in-game bonuses for scanning real-world locations with their phone cameras. Users who opted in uploaded those recordings to Niantic, the company that created the game in partnership with Nintendo.
Those scans — approximately 30 billion individual environmental recordings, according to reports — were used to train Niantic’s foundation AI models to recognise and interpret physical spaces. Players who agreed to the feature unknowingly granted Niantic a licence to resell that imagery to third parties, buried in the terms of service. In 2025, Niantic sold its entire gaming division, including Pokémon Go, to Scopely — a company owned by Saudi Arabia’s Savvy Games Group — for $3.5 billion. The non-gaming operations were spun off into a separate entity called Niantic Spatial Inc.
From augmented reality to autonomous warfare
In December 2025, Niantic Spatial announced a partnership with Vantor, a defence and intelligence firm that specialises in spatial detection software for drones used by military organisations. Vantor was formerly known as Maxar Intelligence and rebranded in October 2025. The deal aims to solve what the partners describe as “a critical vulnerability in modern operations: GPS unavailability, spoofing, interference, and jamming”.
The technology works by combining Niantic Spatial’s Visual Positioning System (VPS) with Vantor’s Raptor visual positioning software. Raptor was originally designed to give aerial drones the ability to navigate without GPS by recognising visual landmarks from cameras. The partnership extends that capability to the ground, creating a shared coordinate system derived from live video feeds alone. In practice, a drone flying into a contested zone where satellite signals are deliberately blocked can look at the terrain below, match it against the vast library of real-world scans fed into the AI model, and pinpoint its location with high precision. Brian McClendon, Niantic Spatial’s chief technology officer and a former lead on Google Maps and Street View, has described the approach as ideal for robots operating in dense urban environments or war zones where signals are either weak or deliberately jammed.
Field testing of the integrated system was planned for early 2026. Peter Wilczynski, Vantor’s chief product officer, told Tectonic Defence in a December interview that “the modern battle space is going to be complete with different systems, and you’re going to want to upgrade those systems quickly — bringing new hardware online faster than new software”.
Both companies have been careful to draw a line between the training data and the current partnership. A spokesperson for Niantic Spatial told Guardian Australia that “AR Scans collected through Pokémon Go were submitted voluntarily by players who opted into the feature and were subject to the applicable Terms of Service and Privacy Policy at the time”. The company stated that the ground scans from the game were not provided to Vantor as part of the agreement, but acknowledged that those scans had been used to train the foundation models that underpin Niantic Spatial’s VPS technology. Vantor itself denied using Pokémon Go data directly for its current military applications but declined to say whether the models it plans to deploy were trained on such data in the past. Other reports indicate that Vantor’s Raptor system is built on Niantic’s models, meaning the Pokémon Go data is, indirectly, already embedded in the system.
Vantor’s military footprint is substantial. In February, the company announced a deal with the US Army worth up to $217 million for training software under the One World Terrain program, which creates a global synthetic environment for mission rehearsal. It has also partnered with BAE Systems to integrate its Raptor software with BAE’s geospatial intelligence tools, enhancing targeting capabilities for drones operating in GPS-denied environments.
Ethical concerns and the ‘product’ problem
The use of civilian smartphone data — gathered through a game beloved by children and adults alike — for military drone navigation has drawn sharp criticism from privacy advocates. Tom Sulston, head of policy at the tech policy think tank Digital Rights Watch, warned that most people do not read the vast legal documents they agree to when playing a video game. “While they may have disclaimers in their Ts&Cs, we know that most people don’t read vast legal documents when they want to play a video game,” he said. “We need regulators to focus on ‘best interests of the user’ or ‘fair and reasonable’ tests to keep users safe from exploitation like this. While we’re waiting for the government to catch up, it’s important that we remember that ‘free’ software services often treat the user not as a customer but as the product to be sold.”
Dr Rob Nicholls, senior researcher at the University of Sydney’s centre for AI, trust and governance, said the case was likely the tip of the iceberg. He pointed to the 2018 Strava heatmap incident, where aggregated fitness tracker data inadvertently revealed the locations of military bases and patrol routes, prompting reviews within the US Department of Defense and changes to Strava’s visibility settings. A subsequent investigation by Le Monde in 2024-25 highlighted that such risks persist. “We have already seen that Strava data has been used to identify the location of military facilities,” Nicholls said. “Indeed, directives not to use devices with GPS and sharing have come from a number of different militaries.”
The broader trend is clear: augmented reality technology — once the preserve of smartphone games — is increasingly being adopted by armed forces for training and operations, with systems like the US Army’s Integrated Visual Augmentation System (IVAS) using AR headsets to overlay real-time data onto the battlefield. AI analysis of training data is also being used to create dynamic, personalised learning modules for soldiers. For the millions of players who spent years scanning their local parks and high streets for in-game rewards, the knowledge that those scans may now help guide military drones into conflict zones represents a jarring collision between the virtual world and the very real ethical dilemmas of modern warfare.
