BadWAM: When World-Action Models Dream Right but Act Wrong
Introduces BadWAM, exposing adversarial vulnerabilities in world-action models (WAMs) that couple future-state prediction with robotic action generation. Small visual perturbations induce 'World-Action Drift Attacks': an action-only variant collapses task success from 96.5% to 43.1%, while a stealthier 'imagination-preserving' variant keeps the model's predicted future looking clean while still forcing harmful actions, showing that moderate regularization can mask the drift entirely.
Why it matters
Reached 25 upvotes on Hugging Face Daily Papers; demonstrates that a model's internal 'imagination' can look correct even as its actions silently diverge, a concrete interpretability/safety failure mode for embodied AI.
Importance: 3/5
HF Daily Paper (25 upvotes) exposing adversarial vulnerability in world-action models.