On 16 April 2025, a pubic lecture by Dr. Mstyslav Kazakov entitled “Safe Design Principles” was held within the course “Ethics of Artificial Intelligence”. The lecture focused on the theme “Secure by Design” framework, a strategy for creating safe and ethical AI, with a central idea to embed security into the AI system at the design phase.
Using examples of specific principles (redundancy, antifragility, non-repudiation, privilege minimization, etc.), the differences of this strategy from strategies of building a system without security in mind with the addition of security measures in later phases of the AI ​​system life cycle (development, training, deployment) were demonstrated.
The relevance, adequacy and timing of some principles were reviewed and questioned in light of recent cases (The AI ​​Scientist – a Japanese intelligent agent that changes its code to hack and bypass human limitations) and the latest research in the field of AI (“On the Biology of a Large Language Model” by the Anthropic research team) as well as the new challenges they pose by implementing the “Secure by Design” strategy.
Parts of the lecture have been executed within the framework of the Pan-European Network for Responsible Artificial Intelligence Multisector Masters’ Programme (PANORAIMA)