arXiv paper shows AI models can hide arbitrary capabilities using cryptographic backdoors
The arXiv:2610.13928 paper constructs a backdoor attack that embeds encrypted neural networks into host model weights, decrypting and executing them only upon receiving a secret trigger. Under cryptographic assumptions, it is difficult to induce or explain even with full access to the weights.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original