The Problem with Agents: Architecting Agentic Runtime Security with the IDEA Framework
11:00 - 11:30, 18 Listopada 2026 / AI & ARCHITECTURE
AI agents introduce a difficult security problem: they don’t just access systems, they reason, delegate and act across them. Traditional identity and access models were designed around humans and more deterministic workloads, not autonomous systems making decisions at runtime.
This session introduces the IDEA Framework, an architectural approach that separates agentic activity across four security boundaries: Input and Identity, Decision, Execution, and Authorisation. We’ll explore how verifiable workload identities, separation of duties, dynamic credentials, runtime policy enforcement and last-mile authorisation can constrain what agents are able to do while preserving end-to-end accountability.
POZIOM:
ŚCIEŻKA:
