Google has pulled back its newly launched artificial intelligence image-generation feature from Google Earth after just days of public availability, citing concerns over misuse for spreading false and manipulated content. The technology giant announced it would temporarily suspend the "create image" tool while developing improved safeguards to prevent potential disinformation, acknowledging that despite legitimate professional applications, bad actors had already begun generating fabricated satellite imagery that violated the company's content policies.
The feature, introduced on July 30, represented an ambitious attempt to integrate Google's Nano Banana 2 image-generation model directly into Google Earth, the company's widely-used mapping platform. Users could zoom into any location worldwide and generate synthetic satellite, aerial, and three-dimensional imagery in seconds by simply describing what they wanted to visualise. Google had framed the tool as empowering professionals to plan real estate developments, explore historical changes, and visualise potential future scenarios — seemingly innocuous use cases for a powerful new capability.
However, the announcement triggered immediate alarm among the global disinformation research community, who recognised the profound risks inherent in combining a trusted geospatial reference platform with generative artificial intelligence. Digital investigations expert Henk Van Ess encapsulated the concern succinctly: Google had spent two decades building a resource the world relied on for verification and fact-checking, only to introduce a feature that could fundamentally undermine that trust by fabricating realistic-looking satellite data.
The vulnerability became starkly apparent when journalists at AFP tested the tool during its brief window of availability. Within hours, the system generated convincing fake satellite imagery depicting catastrophic events with genuine global significance: an explosion flattening central Paris, a nuclear facility in Iran struck by military action, a fresh bomb crater scarring Russian territory, and what appeared to be an Islamic State training compound in Syria. The system also obliged requests to visualise flooding across Bangladesh, a migrant caravan assembly at the United States-Mexico border, and warehouses in Georgia that appeared to validate baseless election fraud allegations.
Brady Africk, a satellite imagery specialist at the American Enterprise Institute, articulated why these synthetic images pose such danger. Convincing fabricated satellite photographs could proliferate rapidly across social media and news networks, bypassing initial scepticism by leveraging decades of public confidence in satellite data. Once such images circulate widely, correcting the record becomes exponentially harder, as the false version has already lodged itself in public consciousness and across countless digital platforms beyond the original distribution source.
The stakes extend beyond mere embarrassment or momentary market jitters. In March 2024, a single artificial intelligence-generated image showing an explosion at the Pentagon was sufficient to create meaningful volatility in financial markets before fact-checkers could intervene. Similarly, early in the United States-Iran conflict, manipulated satellite imagery purporting to show destroyed American military installations accumulated millions of views across social platforms before wider debunking efforts took effect. Each incident demonstrates how synthetic geospatial media can trigger real-world consequences ranging from economic disruption to escalated international tensions.
Google initially attempted to mitigate these risks through technological transparency measures. The company embedded SynthID digital watermarks—invisible markers identifying content as artificially generated—into all images created through the tool. The company also emphasised its formal artificial intelligence policies explicitly prohibiting users from generating content for "misinformation, misrepresentation or misleading activities." However, these safeguards proved demonstrably inadequate. Watermarks remain trivial for sophisticated actors to remove, and policy documents carry limited enforcement weight when bad-faith actors operate across multiple jurisdictions and platforms.
The pushback against Google's feature came from an unusually broad coalition of expertise. The London-based Centre for Information Resilience, which tracks disinformation campaigns globally, warned that satellite imagery trust had required decades to establish but could be destroyed virtually overnight. Ross Burley, the organisation's executive director, characterised the feature as fundamentally irresponsible given what was at stake. GeoConfirmed, a volunteer-staffed open-source intelligence network that verifies visual evidence from active conflicts worldwide, reported observing users already exploiting the tool to misrepresent military positions and mislabel buildings as schools.
GeoConfirmed's statement proved particularly telling: within the tool's brief operational window, the only readily apparent practical application for a substantial portion of users appeared to be precisely what Google's policies forbade—manufacturing and spreading disinformation. This observation cut to the heart of the dilemma: a technology offering genuine utility for professional applications simultaneously provides an exceptionally powerful platform for malicious use, with the malicious applications proving trivially easy to execute at scale.
For Southeast Asia specifically, the implications warrant careful consideration. The region has experienced demonstrable harms from election-related disinformation and cross-border misinformation campaigns. Indonesia, the Philippines, and Malaysia have all confronted challenges where synthetic or manipulated imagery fuelled communal tensions, election interference, or geopolitical escalation. A tool enabling realistic satellite forgery could amplify these existing vulnerabilities, allowing actors to fabricate evidence of border violations, environmental destruction, or military movements to inflame regional tensions.
Google's decision to roll back the feature represents a rare instance of a major technology company implementing preventive guardrails before anticipated harms fully materialised. Rather than permitting widespread deployment and implementing reactive controls after disinformation spread, the company acknowledged legitimate concerns and suspended the tool. The announcement signals that even companies with enormous resources and sophisticated safety teams recognise that certain technological capabilities require substantially stronger protections before public release—a lesson increasingly relevant as generative artificial intelligence capabilities continue advancing rapidly.
The company now faces the genuine challenge of designing safeguards robust enough to prevent deliberate misuse while preserving the tool's beneficial applications for geospatial professionals, urban planners, and researchers. This represents a fundamentally difficult technical and policy problem with no obvious solution. Whether Google can develop effective protections remains uncertain, but the decision to pause reflects growing recognition across the technology sector that releasing powerful generative tools without adequate consideration of disinformation vectors carries unacceptable risks to public discourse and international stability.
