Fake satellite images: how AI undermines photographic evidence once again (Google still kicking itself)
Can fake satellite images strip imagery of its value as evidence? On 30 July 2026, Google added a tool to Google Earth that generated a realistic image of any point on the planet. Less than 24 hours later, the company pulled it. The episode was short, but it raises a question that goes well beyond mapping: what makes an image prove anything?
Key points in 30 seconds
- Google removed the "create image" tool from Google Earth on 31 July 2026, after warnings from disinformation researchers.
- A fake satellite image is especially convincing because it sits on real coordinates, inside a source people use to check facts.
- The risk cuts both ways: believing a fabricated image, and dismissing a genuine one.
- Certificall certifies field photos at the moment of capture: an eIDAS qualified electronic time stamp, cross-checked geolocation and a qualified electronic seal.
Google Earth: an AI-generated imagery tool withdrawn within 24 hours
The "create image" tool produced an artificial image from Google Earth's satellite, aerial and 3D data. Google presented it as a way to visualise history or plan real estate projects.
Testers quickly found other uses. BBC Verify produced a collapsed Eiffel Tower and Russian tanks on the streets of Kyiv. GeoConfirmed, a collaborative project that verifies conflict imagery, reported fake military positions and buildings falsely presented as schools.
Google first pointed out that its images carried invisible digital watermarks. It then rolled the feature back "while we work on implementing stronger guardrails", according to its public statement. Ross Burley, executive director of the Centre for Information Resilience, put the stakes plainly: trust built over decades could be lost in a single day.
Why fake satellite images are more deceptive than an ordinary deepfake
A satellite image is a view of the Earth's surface captured from space. For open-source investigators, it confirms or refutes that an event took place at a given location.
An image generated inside Google Earth inherits that reputation. The terrain, roads and buildings around the invented element are real. Basic checks, such as comparing the scene with a map, therefore reveal nothing unusual.
Fakes of this kind have already had real effects. In 2023, an AI-generated image of an explosion near the Pentagon briefly moved financial markets. Everyday photo tampering techniques follow the same logic: the fake borrows the appearance of the real thing.
The real risk: when no photographic evidence convinces anyone
In 2019, legal scholars Robert Chesney and Danielle Citron described the "liar's dividend". The liar's dividend is the ability to dismiss genuine evidence by claiming it was fabricated.
Brady Africk, a satellite imagery analyst at the American Enterprise Institute, made the same point to AFP. In his view, the spread of fakes erodes public trust in all satellite imagery, including the authentic kind.
Businesses face the same mechanism. In an illustrative scenario, a contractor sends an insurer photos of water damage recorded that same day. The other party objects: "with AI, anyone could have made that". If nothing documents how the photos were taken, the debate turns on how they look. And appearance no longer proves anything.
An ordinary image
Its date and location rely on editable metadata. Nothing shows it has not been edited or generated. Its value depends on how much you trust the person who supplied it.
A photo certified at capture
Its date is guaranteed by a qualified electronic time stamp. Its location is cross-checked when the photo is taken. A qualified electronic seal makes any later change detectable.
What gives an image evidential value: the conditions of capture
Under French law, facts can be proven by any means: Article 1358 of the French Civil Code admits all forms of evidence, and the judge assesses their weight. An ordinary image is therefore admissible, but its credibility rests on its appearance, which generative AI can now imitate. What makes a photo or video reliable evidence is the way it was produced.
The European eIDAS Regulation (No 910/2014) gives these conditions a precise framework. Under Article 41, a qualified electronic time stamp enjoys a presumption of the accuracy of the date and time it indicates, and of the integrity of the data. Under Article 35, a qualified electronic seal enjoys a presumption of the integrity and correct origin of the data.
Geolocation has no legal presumption of its own. Once sealed with the photo, it cannot be altered without the change showing, and its reliability rests on cross-checking location data at the time of capture.
Four questions to ask about any image presented as evidence
- When? Is the date guaranteed by a third party, or merely written into the file?
- Where? Was the location recorded and cross-checked when the photo was taken?
- Who? Do we know which device took the image, and for which organisation?
- Since then? Can we show the image has not been modified after capture?
Certifying the real at the source with a timestamped, geolocated photo
Certificall answers these four questions for field photos and videos. Every timestamped and geolocated photo taken in the app receives a qualified electronic time stamp issued by a qualified trust service provider (QTSP). It also receives an eIDAS qualified electronic seal and a SHA fingerprint, in a PDF/A-3B certificate archived in France.
The scope must stay clear. Certificall does not certify satellite images and does not authenticate an image after the fact: it certifies what is captured in its app. This approach complements tools that detect or label AI-generated content. It is about certifying the real rather than chasing the fake.
With Certificall, the contractor in the water damage scenario no longer argues about how the photos look. The certificate establishes the date, location and integrity of each shot, which is why certified photos for insurance claims hold up better when challenged. The process is detailed on the page presenting our certification technology.
Contact us to integrate Certificall into your business processes
Field photos dated, located and sealed at the moment of capture.
The questions we really get asked about fake satellite images
- ❓ Why did Google withdraw its image generation tool from Google Earth?
- Google saw generated images being shared that appeared to violate its policies, after warnings from disinformation researchers. The tool, launched on 30 July 2026, was withdrawn the next day while stronger guardrails are put in place.
- ❓ How can you spot a fake satellite image?
- There is no reliable visual test. Investigators compare the image with other sources captured around the same date, trace its origin and treat context-free screenshots with caution.
- ❓ Can a photo still serve as evidence in the age of generative AI?
- Yes. Under French law, facts can be proven by any means, but the weight of a photo depends on how it was captured. A qualified electronic time stamp and an eIDAS qualified electronic seal give it a presumption of date, integrity and origin.
- ❓ Is a geotagged photo enough to prove where it was taken?
- Not always. Coordinates stored in a file's metadata are easy to edit. Geolocation that is cross-checked and sealed at the moment of capture is much harder to challenge.
- ❓ Can Certificall certify a satellite image or an existing image?
- No. Certificall certifies photos and videos taken in its app, at the moment of capture. For an uploaded image or document, Certificall provides an analysis and a confidence score, without being able to certify its origin.
