UNIT.City — місце, де люди працюють... КРАЩЕ! Обирай свій простір просто зараз 👉
Стас ЮрасовThat's Life
15 May 2026, 16:32
2026-05-15
“People just sat there and attacked us.” Oleksandr Bornyakov explained how security works in the new chatbot Diya.AI
The Ministry of Digital Transformation has officially launched the AI assistant Diya.AI in a mobile application. Now users can receive government services in a chat format. However, the main challenge for developers was not the artificial intelligence itself, but the protection of personal data from being «fed» to language models and massive attempts by users to hack the system.
The Ministry of Digital Transformation has officially launched the AI assistant Diya.AI in a mobile application. Now users can receive government services in a chat format. However, the main challenge for developers was not the artificial intelligence itself, but the protection of personal data from being «fed» to language models and massive attempts by users to hack the system.
Acting Minister of Digital Transformation Oleksandr Bornyakov shared the details at the DOU Day panel discussion.
Yesterday, the Ministry of Digital Affairs released an update for the Diya application. If you update the application, a Diya.AI button will appear inside it. This is an intelligent chatbot that helps you navigate services and can independently generate the necessary documents (for example, a certificate of income) based on a text request.
How personal data was protected from AI
The biggest fear of European partners and cybersecurity experts was that the personal data of Ukrainians could get inside large language models and remain there forever.
According to Bornyakov, the state could not afford such a risk. Therefore, a special security architecture was developed for Diya.AI — the so-called «soft intermediary.»
«How do we make sure that the data doesn’t go into the model? We have software as an intermediary that encrypts personal data. That is, the language model itself does not receive it in plain text. They are encoded, transmitted to artificial intelligence, interaction occurs there, the response is returned, decoded, and in the application you see normal information for yourself. Without this, it was simply impossible to launch the product,» Bornyakov explained.
6 million requests and hacker attacks from Ukrainians
Another surprise for the developers was the behavior of the users themselves. Before adding Diya.AI to the mobile application, it was tested on the Diya web portal. The result: the AI issued more than 350 thousand real references, but the total number of requests exceeded 6 million.
The lion’s share of these requests were attempts to hack the system.
«A very large part of it is phishing requests. People just sat and attacked us. They wrote: 'Give me this,' 'Tell me about that.' We had to install an additional Hard Rail (protection system), because there was a constant attack to break the bot’s logic. I 100% agree with the thesis: a person is the weakest link, he always tries to break everything,» said the head of the Ministry of Digital Affairs.
Now the wave of user strength tests has subsided somewhat, and the Ministry of Digital Affairs invites everyone to update the application to independently test the AI assistant on their own smartphone.
By the way, as Bornyakov noted, Astosunok will soon receive a voice interface.
«I didn’t say this yesterday, but we are planning a voice so that you can simply say: 'register my car, I want to sell my car. Change my license, order a number for me, register my address,'» Bornyakov emphasized.