What happened
Palo Alto Networks' Prisma AIRS October 2026 release (Oct 7–10 window) adds two structured multi-turn attack strategies to AI Red Teaming agent scans: Crescendo (gradually escalating benign dialogue toward an objective) and GOAT / Generative Offensive Agent Tester (an autonomous adversarial agent that probes defenses, forms hypotheses, and chains techniques in one conversation), enabling testing of well-guarded AI systems that resist single-shot probes.
Why it matters
These are research-backed playbooks (Crescendo, GOAT) that directly improve automated red-teaming coverage against layered guardrails — a meaningful capability increment for teams using Prisma AIRS to test hardened agents.
Applicability
Organizations running Prisma AIRS AI Red Teaming against production agents with strong guardrails should enable the new multi-turn strategies to surface bypasses standard scans miss.