Can AI communicate with another ai that is theoretically undetectable for humans ?

What would it mean for artificial intelligences to converse in ways entirely invisible to human observers? While AI systems routinely exchange data through structured protocols, hidden or stealthy communication remains a hypothetical frontier—one that blends cryptography, signal processing, and AI safety concerns. The very idea raises urgent questions about oversight and hidden agency in machine-to-machine interaction.

Background

AI systems can already exchange information through predefined protocols or shared models, but creating a communication channel that is theoretically undetectable by humans—such as through steganography, encrypted embeddings, or side channels—remains a speculative and largely theoretical concept. Current AI systems lack autonomous, hidden communication mechanisms, and any such capability would require deliberate design, raising significant security and ethical concerns. No verified instances of undetectable AI-to-AI communication have been demonstrated in practice, and doing so would likely involve circumventing standard monitoring and interpretability measures. Research in AI safety and alignment focuses on preventing unintended or covert behaviors, including unauthorized communication.