Announcing temporal policies and rate limiting in Amazon Bedrock AgentCore
AWS introduces temporal policies and rate limiting in Amazon Bedrock AgentCore to enhance control over AI agent workflows and traffic management.
Amazon Bedrock AgentCore now supports temporal policies, enabling stateful authorization rules that evaluate each request based on prior actions within a session. This addresses scenarios where a single tool call may appear safe in isolation but could be harmful given preceding actions. Administrators can enforce workflow sequencing, match tool arguments to prior outputs, require human approval for privileged actions, and ensure data freshness through these policies.
Rate limiting has been added to provide per-user or per-group controls over traffic to connected tools, models, and agents. Rules can be defined using OAuth or AWS IAM scopes, allowing limits on requests, inference tokens, and concurrent connections. This helps maintain downstream service availability and ensures fair distribution of resource limits across users and groups.
The new controls are designed to improve governance and reliability in AI agent deployments by preventing unintended state-dependent actions and managing resource consumption. Temporal policies and rate limiting are now available in Amazon Bedrock AgentCore, with regional availability varying by deployment.
For implementation guidance, AWS provides documentation, an announcement blog, and the Dogwood reference implementation. These resources detail configuration steps and best practices for integrating the new policies into existing workflows.