AWS just shipped the first cloud-native governance layer for AI agent tool calls. And the design is smarter than I expected.
On March 19, 2026, AWS Bedrock AgentCore Policy reached general availability across 13 regions, covering the US, EU, and Asia Pacific. The pitch: intercept every OpenClaw agent tool call, check it against your rules, and block anything that violates policy. All before the tool actually executes.
How It Works
Teams write governance policies in plain English. AgentCore converts them into Cedar, AWS's open-source policy language. Cedar is deterministic, not LLM-based, so enforcement runs at millisecond latency and handles thousands of decisions per second.
The default posture is deny-all. Your OpenClaw agents can't call any tool without explicit permission. That's a big philosophical shift from the "allow everything, scan after" approach most security vendors take.
The AgentCore Gateway sits between your agent and its tools. Every call gets checked. Violations are blocked, logged, and alerted. No exceptions, no "we'll flag it later."
Worth noting: on March 12 AWS also shipped AgentCore Memory Streaming, which sends real-time Kinesis notifications whenever an agent's memory changes. If you're building audit trails for regulated industries, that matters.
The RSAC 2026 Context
This launch didn't happen in a vacuum. RSAC 2026 has been wall-to-wall agent governance announcements. Cisco DefenseClaw does scanning and sandboxing. Microsoft announced Entra Agent ID for identity management. CrowdStrike and Palo Alto both showed agent-specific tooling. And now AWS tackles the policy enforcement angle.
The stat driving all of this? Roughly 63% of organizations say they can't enforce purpose limitations on their AI agents. That's a problem when your agents can book flights, modify databases, or send emails on your behalf.
What This Means If You Use ClawHosters
AgentCore targets enterprise IT departments deploying OpenClaw on AWS infrastructure with Cedar policy language and IAM integration. That's a different world from what we do.
ClawHosters serves individuals and small businesses who want managed OpenClaw without the AWS complexity. We handle security defaults out of the box: container isolation, firewalls, auto-updates, safety scanning. No Cedar files required.
If you're comparing approaches, our self-hosting vs managed guide covers the trade-offs.