AWS Agent Registry hits general availability to govern enterprise AI agents

AWS Agent Registry addresses a fast-growing enterprise anxiety: 'rogue' or untracked agents proliferating inside production systems. Generally available since August 31, the registry automatically detects agents running on Amazon Bedrock AgentCore and pulls them into a single catalog with lifecycle management, cost allocation and audit trails, billed at $0.40 per 1,000 records.

The important caveat, and the crux of community discussion, is that the registry provides discovery and governance visibility — not automatic security enforcement. Organizations still need to build their own policy controls on top; the registry tells you what agents exist and what they cost, but won't stop a misbehaving one. Commentators cited examples like Westpac's five production lending agents to argue that inventory and audit are necessary but insufficient without enforcement layers.

This fits AWS's broader agent-infrastructure push this week, which included AgentCore federated data-access patterns, memory lifecycle policies and a multimodal WhatsApp ordering assistant. AWS also committed $5.3 billion and 50MW of capacity to a Saudi AI Zone by 2028, with HUMAIN's ALLAM Arabic LLM coming to Bedrock. For enterprises scaling from a handful of pilots to dozens of agents, governance tooling is becoming table stakes — but the distinction between 'we can see our agents' and 'we can control our agents' is exactly what security teams will press AWS on next.