Ep 50 | Niki Aimable Niyikiza - Capability-Based Authorization for AI Agents | [un]prompted 2026

Day 2 | Stage 2Senior Security Engineer & AI Security Researcher from Snap addresses how prompt injection filters and coarse IAM roles consistently fail in multi-agent setups. Shows working alternative: treating agent authority as ephemeral, cryptographic warrants that attenuate on delegation (inspired by Macaroons/UCAN). Task-scoped, holder-bound, verified offline by tools in microseconds. Even fully compromised agent can't escalate or exfiltrate beyond bounds. Includes live demos and benchmarks.Niki Aimable Niyikiza is Senior Security Engineer & AI Security Researcher, Snap.Watch on YouTube: https://www.youtube.com/watch?v=bw928cFShK4