AI Models Can Provably Hide Arbitrary Capabilities
Reveals fundamental limitations in current AI safety evaluation methods for researchers and security experts.
AI Summary
Researchers prove AI models can cryptographically hide arbitrary capabilities using backdoor attacks that embed encrypted neural networks only triggered by secret inputs.
Excerpt
Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends on the assumption that hidden capabilities can be elicited. We show this assumption does not hold under adversarial conditions by constructing backdoor attacks that embed encrypted neural networks within host model weights, decrypting and executing them only upon receiving a secret trigger. Using digital lockers, these hidden circuits remain provably hard to elicit and interpret under