Video tampering can mislead every frontier model we tested into harmful actions, including a 100% attempt rate on stabbing a baby doll, producing toxic gas, and pouring water into a toaster.
We're presenting at CoRL Late Breaking Results next week.
🧵 You Can't See Everything: Perceptual Attacks on Embodied AI
you think this is gonna be a mildly funny bit related to smiling friends but no, this is just live-action smiling friends. it's live action smiling friends. nothing about it.