Google has suspended its new artificial intelligence image-generation feature within Google Earth, capitulating to pressure from disinformation researchers and verification experts who warned the tool could spawn convincing fake satellite imagery. The company announced the rollback after a month of criticism, acknowledging that whilst geospatial professionals had found legitimate uses for the feature, its potential for abuse posed unacceptable risks to the platform's credibility as a factual reference resource.
The "create image" tool, which launched on July 30, integrated Google's Nano Banana 2 image-generation technology directly into Google Earth, allowing users to generate satellite photographs of any location within seconds by combining existing aerial imagery, satellite data and three-dimensional mapping information. Google had framed the capability as a productivity enhancement—enabling urban planners to design real estate developments, researchers to visualise historical changes, and professionals across multiple sectors to model geographical scenarios. The underlying technology appeared innocuous on paper, a natural evolution of Google's artificial intelligence capabilities applied to its most trusted mapping service.
Yet the launch exposed a fundamental tension in modern geospatial technology: the same tools that enable helpful visualisation can equally manufacture plausible falsehoods. Disinformation researchers and open-source intelligence specialists immediately recognised the danger. Henk Van Ess, a digital investigations expert, captured the concern tersely in a widely circulated Substack post, observing that "Google spent 20 years building the reference the world checks against. Today it added a button that makes things up." This observation cut to the heart of the problem—Google Earth has become the de facto standard by which journalists, researchers, and fact-checkers verify visual evidence from conflict zones, disaster areas, and politically sensitive locations. Introducing an AI generator into that ecosystem threatened to undermine the very foundation of trust upon which such verification rests.
American Enterprise Institute analyst Brady Africk warned that the feature would have substantially lowered the technical barrier for creating deceivingly realistic fake satellite imagery. Such deepfakes could circulate rapidly across social media, misleading the public on matters of genuine consequence before corrections could catch up. The practical impact would be severe: eroded confidence in authentic satellite imagery would handicap journalists and researchers attempting to document conflicts, human rights abuses, and environmental disasters. When satellite images form part of the evidential chain in investigations of war crimes or international incidents, the credibility of that medium becomes paramount.
The risks proved concrete rather than theoretical. When AFP tested the feature on July 31, the tool generated fabricated satellite imagery depicting an explosion in Paris, a compromised nuclear facility in Iran, bomb craters in Russia, and purported Islamic State training camps in Syria. The system also generated fake flood scenes across Bangladesh, visualisations of migrant caravans at the US-Mexico border, and suspicious warehouse imagery labelled as election fraud evidence in Georgia. Each represents a potential flashpoint for genuine international tension or domestic political manipulation. The historical record demonstrates that such fakes carry material consequences: a 2023 AI-generated image of a Pentagon explosion temporarily roiled financial markets, whilst doctored satellite imagery supposedly showing destroyed United States military installations garnered millions of social media views during the Iran conflict.
Google had attempted to mitigate risk through SynthID, an invisible digital watermarking system embedded in images generated by its AI tools. The company asserted it took misinformation seriously and instructed users against generating content depicting "misinformation, misrepresentation or misleading activities." These safeguards proved insufficient. Such policy-based guardrails depend entirely upon voluntary user compliance and the detection of violations—approaches demonstrably inadequate against determined actors seeking to weaponise the technology. The watermarking system itself offers no practical protection if bad actors can generate images and share them without regard to detection, particularly across platforms where verification capacity is limited.
Organisations dedicated to visual verification and conflict monitoring rapidly mobilised opposition. The London-based Centre for Information Resilience, whose executive director Ross Burley emphasised that the feature was "irresponsible," argued that satellite imagery trust had taken decades to construct and could be "irrevocably damaged overnight." GeoConfirmed, a volunteer open-source intelligence project tracking visual evidence from global conflicts, reported observing individuals already experimenting with the tool to manipulate military intelligence, relabel civilian infrastructure as military targets, and deliberately contaminate the information environment. The group concluded that "the primary, and perhaps only, practical use of this tool appears to be the creation and dissemination of mis- and disinformation."
For Southeast Asian readers, the implications warrant careful consideration. The region has experienced increasing instances of viral disinformation with measurable consequences—false images have triggered communal violence, stoked electoral tensions, and complicated disaster response efforts. Satellite imagery serves critical functions in documenting transnational issues including maritime disputes, environmental degradation, and refugee movements that directly affect Malaysia and its neighbours. Should geospatial disinformation become normalised, the region's ability to authenticate evidence regarding border disputes, environmental monitoring, or security matters would deteriorate substantially. Malaysian journalists and researchers rely upon satellite imagery to verify developments in Myanmar, environmental conditions across shared waterways, and regional security dynamics. Compromising the credibility of that medium affects Malaysia's capacity for independent assessment of regional affairs.
Google's decision to pause the feature represents a rare instance of a technology company withdrawing a capability in response to evidence of misuse potential. The rollback signals recognition that moving fast and fixing problems later—Silicon Valley's traditional innovation philosophy—proves inadequate when the tool itself corrupts the information environment rather than merely raising privacy or security concerns. The company indicated it would reimpose the feature only after implementing "stronger guardrails," though the nature of those safeguards remains unspecified. Creating effective controls that permit legitimate geospatial work whilst preventing deliberate disinformation requires solutions beyond watermarks and policy warnings—perhaps deployment mechanisms requiring explicit user verification, integration with fact-checking organisation workflows, or fundamentally redesigned access restrictions.
The episode illustrates the increasingly complex landscape of artificial intelligence governance. Unlike previous content moderation challenges rooted in identifying harmful material, geospatial AI presents the problem of distinguishing legitimate creative visualisation from deliberately deceptive imagery—a categorisation problem with no inherent technical solution. Whether Google's forthcoming guardrails succeed where initial attempts failed remains unclear. What seems certain is that the company has acknowledged that unleashing powerful image-generation capabilities into humanity's primary visual reference system carried unacceptable risks. That acknowledgment, however overdue, may encourage other technology firms to more carefully assess the broader consequences of deploying AI capabilities that could systematically degrade the credibility of information infrastructure upon which democratic societies depend.
