Policy enforcement now ships at build time across Bedrock AgentCore, Vertex, Azure AI Foundry, and LangChain. The exploit that broke OpenClaw never touched the model at all.
Build-time policy answers what the agent is allowed to do. It has no opinion on why it just did that." That's the whole gap in one sentence.
The OpenClaw chain is the uncomfortable part. The attacker never touched the model, they just grabbed the policy engine through a stolen token and a WebSocket server that never checked origins. The sandbox was built to contain a hijacked model, not survive someone rewriting the rules from outside.
Build-time enforcement is the floor. Runtime behavioral monitoring is the house. Most teams are still pouring the foundation and calling it done.
cisco’s Agent Runtime SDK is a build-time layer that wires policy enforcement directly into an agent’s code, before it ever ships.
pretty cool imo. AMA if this can help
you out.
Build-time policy answers what the agent is allowed to do. It has no opinion on why it just did that." That's the whole gap in one sentence.
The OpenClaw chain is the uncomfortable part. The attacker never touched the model, they just grabbed the policy engine through a stolen token and a WebSocket server that never checked origins. The sandbox was built to contain a hijacked model, not survive someone rewriting the rules from outside.
Build-time enforcement is the floor. Runtime behavioral monitoring is the house. Most teams are still pouring the foundation and calling it done.
"floor vs house" is a cleaner way to say it than i managed. nice work.