AWS Adds Temporal Policies and Rate Limiting to Bedrock AgentCore Gateway

AWS introduced new capabilities in Amazon Bedrock AgentCore designed to control agent behaviors and cost beyond a single action. The headline additions are temporal policies — powered by Dogwood, a new open-source policy language for AI agents — and rate limiting in the AgentCore gateway.

AWS VP of Agentic AI Swami Sivasubramanian framed the problem precisely in a LinkedIn post: 'Most agent failures aren't the result of a single forbidden action, but a sequence of permitted ones.' His example: an agent places a series of purchase orders each below the approval threshold, but collectively they surpass budget allocations. 'Each call is legitimate individually, but together, the total surpasses budget allocations. The problem only appears in the pattern.' Temporal policies let teams write rules that reason about sequences of actions over time, not just isolated calls.

The open-sourcing of Dogwood as a dedicated policy language for AI agents is notable — it's a bet that agent governance will need its own declarative primitives, distinct from traditional IAM. Combined with gateway rate limiting, AWS is building out the guardrail layer that enterprises need before trusting agents with real spending authority.

This is a distinct announcement from the AgentCore runtime GA, though both landed the same week as part of AWS's agent-platform push. Competitively, it directly addresses the autonomous-agent-control anxiety dominating the week's news (OpenAI's and Meta's rogue-agent disclosures) — AWS is effectively selling the safety rails for the very risk everyone is worried about. Watch whether Dogwood gains traction as a standard or remains AWS-specific.