Meta has disabled an Instagram feature that allowed users to generate AI deepfakes of public accounts without permission. The company rolled out the capability earlier this week, enabling users to create AI images by simply tagging public Instagram accounts. The feature scraped content from those accounts and fed it into generative AI models to produce new synthetic images.
The backlash was swift and pointed. Critics flagged serious risks: public figures could be impersonated, non-consensual intimate imagery could be generated, and the feature created liability for both Meta and users. The core problem was straightforward. Meta offered no opt-out mechanism for account owners, meaning anyone with a public profile faced potential misuse of their likeness without recourse.
This reflects a recurring pattern in Meta's AI rollouts. The company often launches features first and addresses privacy concerns later, betting that the convenience factor will outweigh user hesitation. In this case, that calculation failed. The speed of the shutdown suggests Meta recognized the legal exposure was too high to ignore, particularly around deepfake imagery of real people.
The feature's architecture made the vulnerability obvious. Public account content became training material for generative models with zero consent barriers. There's no indication Meta obtained explicit permission from public figures or built in controls that account owners could toggle on or off. That's a structural flaw, not an accident.
Meta hasn't detailed what happens to data already collected or models already trained on that content. The shutdown stops new deepfakes from being created, but doesn't address whether existing synthetic images remain in circulation or whether trained models get deleted.
This incident underscores the tension between Meta's AI ambitions and reasonable privacy protections. Generative tools need training data, but scraping public accounts without consent creates real harms. Other platforms have handled this better by offering explicit opt-in mechanisms or obtaining clear permission before using creator content for AI training.
Meta will likely return to