Exploring the context of online images with Backstory

Google DeepMind has a new experimental tool called Backstory, and it's aimed at a problem that's gotten messier as generative AI has flooded the internet with images: figuring out whether what you're looking at is trustworthy. Built on Gemini, Backstory takes an image and a prompt, then digs into whether the picture was AI-made, where it's shown up online before, and whether it's been edited or repurposed since it first appeared.

The key insight driving this project, according to researchers Avneesh Sud and Chris Bregler, is that spotting an AI-generated image and knowing whether it's misleading are two different jobs. A real photo can be cropped, relabeled, or dropped into a new context to tell a false story. Meanwhile, an AI-generated image might be perfectly honest — illustrating a fictional scene or supporting a factual point without pretending to be a photograph. Backstory tries to close that gap by pairing detection tech with a broader look at an image's history: metadata, prior online appearances, how its framing has shifted over time.

The output isn't just a yes-or-no verdict. Users get a readable report and can keep asking follow-up questions, letting the tool explain how an image's story evolved rather than just slapping a label on it. That conversational layer matters, because context rarely fits into a single sentence.

Google DeepMind is treating this as a slow rollout rather than a finished product. The company is testing Backstory with a group that includes journalists, content creators, and information specialists, and says it plans to gather feedback throughout the year before pushing the tool further. It's also framing this as part of a bigger, multi-year effort — last year's research on trustworthiness and provenance laid some of the groundwork — and one that it argues can't be solved by any single company. Governments, academics, and civil society groups all get a mention as necessary partners in keeping the internet's information ecosystem from completely falling apart.