Mosseri suggests labeling what a real camera captured instead of labeling AI
Instagram head Adam Mosseri says detection of AI may fade, so marking camera-captured posts could be more practical.
Lenny's Podcast: Product | Growth | Career · The rise of taste, human authenticity and judgment in an AI world | Adam Mosseri (Head of IG)Adam Mosseri, the head of Instagram, was asked how Instagram should handle AI posts mixed in with everyone else's posts. He said Instagram should not judge a post by the tool that made it. As Mosseri put it, it should be judged on “the content, the point of view, the person behind the content.”
When asked whether people would want to separate AI from the rest of the feed, he offered a different way to mark things. He said it might be more practical over the long term to label camera-captured content, meaning basically non-AI content, rather than trying to label all AI content. He pointed to two reasons: detection may fail, and context about the account matters.
Mosseri's reasoning on detection is that spotting AI may stop working as models improve. He said Instagram can detect AI-made content right now, and thinks users should be told when something was created with AI. But as models get better, he said, “we might lose the ability to detect that.”
Under that possible approach, the focus would move to identifying work that came straight from a camera, instead of trying to catch every synthetic image. For account context, Mosseri said Instagram should tell people more about who posted something, such as the account's location, profile history, and age, so users can decide whether to trust it.
He pointed to spam accounts as a new problem that makes that context matter. Mosseri described fake accounts using AI-generated imagery, including what he called an AI monk, to sell bogus supplements. The images were not obviously AI, he said, and played on a familiar look or stereotype.
He thinks users should get information and judge posts themselves.
Questions people ask
How would a phone or app prove that a photo was really captured by a camera?
A photo can carry a secure record of where it came from, using a standard called Content Credentials. Newer cameras and phones can cryptographically sign an image at capture, and editing software records changes, so an app like Instagram can show a verified capture history. If that history is missing or broken, there is no proof it came from a camera.
Can Instagram currently tell when something was made with AI?
Yes, in limited ways. Meta attaches labels when its systems detect industry markers, when someone uses Meta's own AI tools, or when a user discloses AI use. The problem Mosseri points to is material that has no marker and looks increasingly photographic, which automated detectors miss.
What happens to a real photo that is later edited with AI?
Not always, and that is part of Mosseri's point. A real photo that is heavily edited or recreated by AI may lose its camera signature, while a lightly retouched photo may keep it. That is why proposals usually show degrees of confidence or edit history, rather than a simple real-or-fake badge.