AI agents and knowledge bases are the new attack target

At the ELCORE DAY 2026 conference, held in Almaty on September 22, the main AI risk for Kazakhstani business was described not as outside hacking tools but as the AI infrastructure companies are busy building right now. GPU clusters, corporate knowledge bases for chatbots and AI agents with access to internal systems have created a new attack surface.

Older security methods cope with it poorly. They were designed for familiar servers, networks and workstations, not for a system that reads documents on its own and calls into accounting tools. A chatbot's knowledge base pulls contracts, prices and correspondence into one place, and an agent with broad permissions becomes a single entry point to all of that data at once. The faster a company connects AI to its 1C, CRM and email, the more it matters to know exactly what that AI can see and do.

Check three things: what data has already ended up in your bot's knowledge base, what permissions your AI agents hold, and whether their actions are logged. A sound benchmark is permissions by job role: in 1Cifer, the accountant's agent sees the accountant's data and the manager's agent sees the manager's, with access mirroring the company structure.