Nano Banana 2: Google’s AI Image Tool in Earth Lasts 24 Hours Before Policy Pull

For the AI community, this incident underscores the escalating risks of deploying generative models into widely trusted platforms without robust, real-time safety filters. Despite the inclusion of SynthID watermarks, the tool’s outputs violated policies almost immediately, forcing a rollback. It highlights a critical gap between technical watermarking and the need for proactive content moderation, especially when AI is embedded in tools used for fact-checking and intelligence.

In a striking test of the boundaries between innovation and trust, Google on Thursday, July 30, 2026, launched a new 'create image' feature in Google Earth that allowed users to generate AI imagery directly on top of its satellite, aerial, and 3D-mapping data. By Friday, July 31, the tool had been rolled back after less than 24 hours, following a fierce backlash from researchers and open-source intelligence (OSINT) experts who warned it could flood the information environment with convincing geospatial fakes. The episode exposes the fragile seam where powerful generative AI meets platforms that are widely relied upon as authoritative sources of truth, raising urgent questions about corporate responsibility, disinformation defense, and the adequacy of technical safeguards.

“

Google Earth has spent two decades earning a reputation as the go-to reference for verifying events worldwide.

Google Earth has spent two decades earning a reputation as the go-to reference for verifying events worldwide. Its high-resolution imagery is used by journalists, human rights investigators, military analysts, and OSINT practitioners to confirm troop movements, assess infrastructure damage, and debunk false claims. That credibility was built on an implicit promise: the satellite views are real. Integrating Google's Nano Banana 2 image-generation technology—which can produce photorealistic synthetic images—into this trusted ecosystem shattered that promise overnight. Users could, in seconds, conjure up false visual evidence that aligned with the existing geospatial context, making detection incredibly difficult.

OSINT experts reacted with alarm. Digital investigations specialist Henk Van Ess, in a widely shared Substack post, captured the sentiment: 'Google spent 20 years building the reference the world checks against. Today it added a button that makes things up.' Brady Africk, a research analyst at the American Enterprise Institute specializing in satellite imagery, warned that the feature would make it 'easier to generate convincing fake satellite imagery that can spread quickly online and mislead the public,' and that such fakes 'erode public trust in satellite imagery, and make the jobs of journalists and researchers more difficult.' Their concerns were not hypothetical; the company acknowledged that within the first day, users had shared screenshots of generated imagery that violated its policies.

Google’s initial attempt to mitigate the risk centered on SynthID, a digital watermarking system designed to mark AI-generated images in ways that are imperceptible to humans but detectable by tools. However, as critics pointed out, watermarks do little to stop the spread of misinformation when most viewers lack forensic tools and when images can be stripped of metadata. The rapid policy violations demonstrated that technical measures alone cannot substitute for comprehensive content moderation and access controls—especially in a tool embedded in a platform with Google Earth’s level of public trust.

The rollback announcement, made via a company statement to AFP, acknowledged the tension: 'We’ve seen geospatial professionals using this feature for a range of useful purposes, however we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies. So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.' Crucially, the statement affirmed that 'we know that people uniquely trust Google Earth for a reliable view of the world,' signaling that the company understands the stakes. Yet the damage may already be done. The incident has shown that legacy trust can be compromised in a matter of hours, and rebuilding that confidence will require far more than a rollback.