1 articles with this tag
New Verifiable Latent Alignments (VLA) framework enables monitoring and steering of hidden AI agent communication channels, mitigating covert collusion.
