AgentKernel: The Trust-Native Agentic Operating System

AgentKernel is a trust-native agentic operating system designed to address the attack surface created when AI agents cross trust boundaries by ingesting untrusted content, combining it with privileged instructions, persisting intermediate beliefs in long-term memory, and invoking privileged tools. The system targets the risk that malicious payloads entering through model inputs can compromise agent behavior.

Modern AI agents routinely cross trust boundaries: they ingest untrusted content, combine it with privileged instructions, persist intermediate beliefs in long-term memory, and invoke privileged tools. This creates an attack surface in which malicious payloads can enter through model inputs and caus