In an age where misinformation can spread rapidly, the advent of artificial intelligence has significantly impacted how visual information is produced and perceived. Recent events, particularly around military actions, have highlighted the challenges posed by AI in manipulating imagery, raising concerns about the accuracy of what we see online.
On February 28, as U.S. military strikes began in Iran, a Chinese AI startup shared a colorized image of a burning ship at Iran’s Konarak Naval Base on the social media platform X. This image closely echoed a black-and-white photograph released earlier that day by Vantor’s News Bureau, raising eyebrows about authenticity.
“They applied an AI program to colorize it,” Stephen Wood, Senior Director of Vantor News Bureau, revealed to SpaceNews. “It was the first time I witnessed someone digitally alter one of our images without consent.”
Meanwhile, the Tehran Times, an English-language daily, circulated manipulated satellite imagery purporting to show the destruction of a U.S. radar base in Qatar. In reality, the image stemmed from an AI-enhanced Google Earth picture of a U.S. base located in Bahrain, as noted by the Central European Media Digital Observatory.
“During the recent events related to Iran, we’ve already observed AI-generated or altered ‘satellite’ images being disseminated to manipulate public information,” commented Bo Zhao, a geography professor at the University of Washington, via email. “This trend is likely to escalate, especially with the capabilities of systems like ChatGPT in generating realistic ‘remote sensing’ imagery. This represents a significant shift from earlier methodologies.”
The emergence of falsified satellite imagery is a relatively novel development.
“While I’m not suggesting it has never occurred, it certainly hasn’t become a widespread practice,” explained Frank Backes, President of the Space Information Sharing and Analysis Center and former CEO of Capella Space.
Executives within the geospatial industry emphasize that the abundance of available satellite images makes debunking falsehoods more manageable than ever before. For example, Vantor’s News Bureau identified AI-generated smoke and fire in an image used in reports about a militant attack on an airport in Niger on January 29. Vantor was able to confirm the image’s inauthenticity as their GeoEye-1 satellite had captured actual imagery of the same airport that day.
“We analyzed the ground features, compared them to the online images, and quickly identified alterations,” Wood stated. “Not only was it not even the correct airport, but the extensive availability of commercial satellite data simplifies the process of highlighting fraudulent images.”
Clients also rely on Ursa Space Systems for cross-referencing different imagery and data types, including electro-optical, synthetic aperture radar, and open-source intelligence, to ascertain what is occurring in a given area, according to Adam Maher, the company’s CEO. “With the commercial availability of data today, you can verify imagery using a secondary source.”
Chain of Custody
Prominent satellite imagery providers prioritize their credibility by implementing strict controls over their supply chains and cybersecurity measures, especially when fulfilling government contracts.
“When established companies release an image, there’s a significantly higher level of confidence in its legitimacy and originality,” Wood noted. “Conversely, materials from lesser-known sources carry a higher risk of alteration.”
This underlines the importance of relying on recognized vendors, as emphasized by Luke Fischer, co-founder and CEO of SkiFi. “It’s no longer enough to trust what you see. Verifying the source is essential.”
Experienced Earth-observation satellite operators offer imagery along with a comprehensive metadata package detailing the sensor used, date and time of capture, the specific location, and the chain of custody, marking every step from satellite tasking to the delivery of data.
When the chain of custody is well-maintained, “you can be assured that no external parties have interfered with an image,” Wood explained.
To safeguard this chain, customers are barred from reselling imagery acquired from recognized vendors or through SkyFi’s geospatial marketplace, which features data from over 50 providers.
Although AI poses challenges for generating misinformation, it simultaneously accelerates the detection of AI-generated artifacts in images while facilitating comparative analysis of images from different providers.
“If an image claims to originate from a specific satellite at a defined time, it can now be validated against actual location data,” Fischer pointed out. “AI offers net benefits in this context.”
“AI is both the challenge and the solution,” stated Zhao, lead author of the 2020 study “Deep Fake Geography” published in the journal Cartography and Geographic Information Science. “We are entering a familiar technological arms race where generation techniques advance, and detection methods strive to keep pace but often fall behind.
“We are shifting from a world of ‘seeing is believing’ to a more precarious landscape where visual evidence doesn’t guarantee belief. Once synthetic imagery reaches a certain level of realism, conventional methods based on visual artifacts or statistical analyses might lose their reliability,” Zhao warned.
Countries may be transitioning toward a “post-truth information environment” where the real issue extends beyond the authenticity of images to how societies handle the inherent uncertainty and visually mediated proof, Zhao concluded.
In this evolving landscape, imagery analysis must combine technical detection of AI artifacts with meticulous tracking of data origins and, critically, enhancing public understanding of AI’s role. “Ultimately, the challenge presented is not solely technical but societal,” Zhao emphasized.
