Ive seen literally one labeled image in a year of scrolling. either i have weird feeds or this isnt being applied. anyone seeing it more often
I see them occasionally on specific high-profile accounts but they’re rare. the auto-detection seems to be very conservative.
Meta’s auto-detection of AI is intentionally cautious to avoid false positives. they prefer to miss synthetic content than to mislabel a real photo. the result is what you describe: minimal labeling in practice.
Yeah makes sense, but it makes the whole feature feel pointless from the user side
The platform-level disclosure approach is going to be reactive forever. user-level disclosure has the same nobody-actually-does-it problem. content credentials at the camera/generator level is the only durable answer
Meta’s classifiers are calibrated to err on the side of not labeling. the public position is ‘avoiding false labels’. the internal reality is ‘we havent figured out how to label at scale without massive moderation overhead’. either reading gets you to the same outcome.
the asymmetry is the real story. the cost to Meta of labeling something incorrectly real-as-AI is high (PR backlash, user complaints). the cost of missing AI content is essentially zero to them. so the system naturally biases to under-labeling.