On the AI Agents Audit Model
О модели аудита ИИ-агентов
2026-03-23
SCID: 54.1/sgvefcgn
Discuss with AI
AI agentsAI systems auditformal verificationlarge language modelstrust in artificial intelligence
Figures from the paper
Abstract (AI)
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.
Key Findings
1
AI agents combine generative models with traditional software and are emerging as a potential replacement programming paradigm.
2
AI-agent development is progressing faster than large language model development, increasing the urgency of evaluating these systems.
3
AI-system auditing is presented as a practical means to increase confidence by verifying that available performance-assurance measures have been considered and implemented.
4
Changing the programming paradigm does not resolve fundamental trust concerns about the reliability of AI-generated results.
5
Rigorous formal proofs of functionality are currently lacking for large language models and, consequently, for AI agents.
Research Object
AI agents
Research Subject
audit model for assessing the trustworthiness and functionality of AI agents in the absence of rigorous formal proofs
Publication Details
Publication Date
2026-03-23
Journal
Publisher
ISSN
Cited by
0
Access Type
Author Information
Download PDF
Subscribe to digest