AI under control: UNN expert summarises the outcomes of the UN summit on technology governance

The first UN Global Dialogue on AI Governance was held in Geneva from July 7 to 10. The event brought together more than 1,000 experts and representatives from the AI sector, over 50 UN partner organisations, and national delegations.

Nikolai Zolotykh, Director of the UNN Institute of Information Technology, Mathematics and Mechanics, and Director of the UNN Artificial Intelligence Centre, was part of the official Russian delegation.

The dialogue was launched by UN Secretary-General António Guterres. In his address, he highlighted three major challenges associated with AI development:

- speed — technologies are advancing faster than we can grasp them;

- concentration of power — resources and influence are held by a limited number of countries and corporations;

- erosion of trust in truth — AI makes it harder to tell reality from fiction.

Joshua Bengio, co-chair of the independent international scientific group studying AI, emphasised that currently there are no effective mechanisms for managing AI at either the national or international level, and the question of fair distribution of AI benefits remains open.

As part of the forum, Russia introduced the sole national session focused on responsible and trustworthy AI. The main topics of discussion were the boundaries of trust in technology and what tasks we can and cannot entrust to machines.

The discussion agenda covered the development of open AI models as a driver of progress, as well as trusted intelligence as the outsourcing of tasks, but not the outsourcing of responsibility. Following the event, it can be noted that the interest of UN countries in Russia's experience with trusted AI is becoming more concrete, focusing on model verification, digital watermarks, and open repositories.

Nikolai Zolotykh also took part in the 16th Academic Conference of the International Telecommunication Union titled "Kaleidoscope - AI and Advanced Technologies for Good." During the sessions "Making Responsible AI Operational: Governance, Standards and Trust" and "AI for Health: Scaling Impact Through Standards, Investment and IP," a discussion took place on moving from ethical declarations to actual mechanisms for controlling AI, as well as on opportunities to overcome barriers to scaling AI in professional fields, such as healthcare.

"When it comes to trust in AI, it’s not about emotions or beliefs—it’s essentially an engineering task. The effectiveness of AI depends on the precision of the algorithms and tasks defined. Trust in AI is built on three pillars: security, explainability, and transparency of AI, as well as human oversight of its outcomes. To achieve this, we need external controls, such as compatible international standards, independent audits and certifications of systems, real-time monitoring of models, and agile management that constantly adapts to bridge the gap between the speed of AI development and regulatory responses. The principles underlying the trusted AI cloud platform we are developing are consistent with many of the points raised at the forum," noted Nikolai Zolotykh.

The trusted AI hardware and software platform for healthcare, being developed at Lobachevsky University, solves a number of important problems of AI application in medicine and is based on the following principles:

- a new systemic definition of trusted AI based on the principles of human control, clinical and information security, explainability and transparency;

- implementing clinical recommendations in the form of a trusted AI model that controls their implementation and controlled by the expert community and the Ministry of Health;

- integration of trusted AI into the healthcare infrastructure and one hundred percent implementation of clinical recommendations in practice;

- ensuring continuous self-learning of clinical recommendations based on language models and the healthcare system with the involvement of key stakeholders;

- harmonious, mutually developing trusted relationships between thepatient, the doctor and AI;

- accessibility of medical services and data on the health status of patients and the population as a whole.