On the AI Agents Audit Model

The topic of AI agents is one of the hottest topics in the development of Artificial Intelligence. Agents are designed to combine generative models with traditional software. The underlying generative models in this combination represent Artificial Intelligence, which must plan solutions to given problems, abandoning rigid algorithmic models. This field of AI agents is developing at a rate that even exceeds the development of large language models. AI agents are already being considered a new replacement programming paradigm. However, issues of trust in Artificial Intelligence systems, which, in reality, represent trust in the results of such systems, are not resolved simply due to a shift in the paradigm of use. Rigorous formal proofs of functionality for large language models and, consequently, for AI agents are lacking. An audit of artificial intelligence systems is a practical way to ensure, in the absence of formal evidence, that all currently available steps to increase confidence in the system's performance have been foreseen and implemented. - from our new paper