Amazon Unveils AI Agent Trustworthiness Framework at VB Transform

AI agents are growing more capable of autonomous business tasks, but IT leaders remain wary of granting them access to enterprise systems. A key challenge is how reliability is measured, as standard EVAL scores offer static performance snapshots that miss real-world predictability. Amazon's AGI Autonomy research lab is moving beyond raw benchmarks, building a framework around consistency, robustness, predictability, and safety. Director Bryan Silverthorn will present the approach at VB Transform 2026, highlighting decoupled systems like sandboxed environments where agents propose changes before execution.