Google Earthの「AIディープフェイク」機能、わずか1日で廃止Google Earth’s AI deepfake tool only lasted one day

Google has withdrawn an artificial intelligence feature from Google Earth less than a day after making it available, following concerns that the tool made it trivially easy to fabricate convincing images of real-world locations. The feature, which launched on Thursday, allowed users to edit satellite imagery using natural-language text prompts, and its rapid removal highlights the tension technology companies face as they weave generative AI into mapping products that many people treat as authoritative records of the physical world.

According to The Verge, the tool essentially let users generate what amount to AI deepfakes of the real world. A person could type an instruction and have the software alter a satellite view accordingly, adding, removing, or transforming elements of a scene that most users assume reflects genuine, camera-captured geography. Because Google Earth is widely regarded as a reliable reference for what places actually look like, the ability to seamlessly rewrite that imagery raised immediate questions about trust and potential misuse.

The risks were quickly demonstrated by researchers and journalists. Henk van Ess of Digital Digging, an outlet focused on open-source investigation techniques, intentionally tested the feature to show how easily it could be used to produce misleading depictions of real places. Such demonstrations are common in the open-source intelligence community, where practitioners rely on satellite and aerial imagery to verify events, track construction, and corroborate claims. A tool that can alter that imagery on demand threatens the evidentiary value that makes those images useful in the first place.

Google's decision to pull the feature so quickly appears to reflect the difficulty of anticipating how generative editing tools will be used once they reach the public. Content moderation for imagery is already challenging, and the stakes rise sharply when the images in question are presented as factual geographic records rather than creative works.

The underlying capability is closely related to the generative fill and inpainting features that have become common across consumer software. Adobe's Photoshop and Firefly, along with Google's own image tools built on its Imagen and Gemini models, let users add or remove objects from photographs using text instructions. Applying that same approach to satellite imagery is technically straightforward, but the context changes the stakes: editing a personal photo is a creative act, while editing a map that others treat as factual can distort a shared reference source.

Google has invested in provenance and detection technologies, including its SynthID system for watermarking AI-generated content, and the broader industry has backed initiatives such as the Coalition for Content Provenance and Authenticity to label synthetic media. It is not clear from the available information whether the withdrawn Google Earth feature included such safeguards, or whether their absence contributed to the decision to shut it down.

The episode also fits a wider pattern of scrutiny around deepfakes and manipulated media. Governments and platforms have grown increasingly wary of synthetic content that can be used for fraud, harassment, or disinformation, and geographic imagery carries particular sensitivity because it can be linked to security, conflict monitoring, and disaster response. Fabricated satellite views could, in principle, be used to invent damage, conceal activity, or mislead during fast-moving events, which is likely part of why the reaction was so swift.

For Google, the quick reversal suggests the company is willing to retract a launch when the potential for harm becomes apparent, even at the cost of an awkward about-face so soon after release. It remains unclear whether the feature will return in a modified form with additional guardrails, or whether Google will abandon the idea of AI-driven satellite editing altogether. Based on the reporting available, the company has not offered a detailed explanation of its reasoning or a timeline for any relaunch.

The incident serves as a reminder that generative AI features which seem innocuous or even playful in one context can raise serious concerns when applied to tools people depend on to understand the real world. As companies race to embed these capabilities across their products, the Google Earth reversal illustrates how quickly a well-intentioned launch can collide with questions of authenticity and public trust.