In brief: Google withdrew its AI image feature in Google Earth, introduced only on July 30, after 24 hours, after it was found capable of generating realistic-looking but entirely fabricated scenes tied to real coordinates.
Google disabled the “Create Image” feature in Google Earth, launched on July 30, within 24 hours after it emerged that the tool could produce fabricated scenes linked to real coordinates. For security leaders, the incident shows how quickly generative AI features in geospatial applications can turn into disinformation risks.
On July 30, 2025, Google integrated its image model Nano Banana 2 directly into Google Earth. Through the “Create Image” feature, users could modify satellite, aerial, and 3D views of real locations via text prompts, for example to visualize construction projects or urban development. Already the following day, it became apparent that the feature could also be used to generate entirely fabricated scenes, which appeared especially credible due to their connection with real map data at genuine coordinates. Researcher Henk van Ess demonstrated this, among other things, using fictional scenes near the US-Mexico border and in connection with an alleged facility in Iran. In its own tests, US broadcaster NPR generated a non-existent fire on an Iranian island as well as a fictional flood at the US Capitol; none of the tested requests were rejected by the system.
For security leaders, the case is relevant because it highlights a recurring pattern in the rapid integration of generative models into established, trusted platforms: linking AI-generated content with real, verifiable map data significantly reduces the detectability of fakes and increases the potential for misuse in disinformation campaigns, for instance in the context of geopolitical tensions or critical infrastructure. The fact that screenshots of generated images could be shared without restriction also illustrates that technical safeguards such as watermarks alone cannot prevent misuse once content leaves the application’s controlled environment.
A Google spokesperson stated that while meaningful use by geospatial professionals had been observed, there were also cases in which shared screenshots appeared to violate the company’s own policies. The feature is therefore being withdrawn while stronger safeguards are developed. Google emphasized that the generated images had been watermarked and never appeared in the public Google Earth view, but were only visible to the respective user. The incident joins a series of similar withdrawals: on July 7, Meta introduced Muse Image, a feature for generating AI images based on public Instagram accounts, and removed it again within a week.
For practical purposes, this means: before generative image or video features are introduced in enterprise applications related to geospatial data, map material, or other sources perceived as reliable, misuse scenarios concerning disinformation and the verifiability of outputs should be systematically tested — not only after rollout.
Source: www.it-daily.net · Published August 3, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.