AI security trilemma for agents

Schneier and Raghavan argue agentic AI faces an 'AI security trilemma': you can be fast and smart, or smart and secure, or fast and secure—but not all three at once. Because agents ingest untrusted data, wield tools, and act in adversarial environments, integrity must be engineered into the architecture rather than bolted on. — This frames AI safety as a foundational design choice that should guide standards, procurement, and regulation for agent systems.