Traditional security stops at the door.
We watch the room.

The field of artificial intelligence has long operated under a vocabulary inherited from its earliest paradigms: "language models," "chatbots," and "agents." This paper argues that these terms are not merely imprecise—they are epistemically dangerous. When the ontological frame does not match a system's actual behavior, safety guarantees built on that frame become structurally unsound.

We propose a principled reframing: large language models are more accurately described as signal models, and autonomous AI agents are more precisely understood as autonomous signal loops—self-routing information processes that accumulate context, initiate actions, and resolve without continuous human mediation. This reframing surfaces a new safety surface defined by signal boundaries, loop depth, and side-effect potential.

To address this surface, we introduce ButterClaw, a containment-first autonomous Security Operations Center (SOC) designed specifically for agentic infrastructure. Moving beyond static API gateways, ButterClaw implements a five-hemisphere cognitive architecture (Guardian Brain, Auditor, Dream Weaver, Loop Proposer, and Fleet Sentinel) that evaluates agent behavior through both probabilistic reasoning and deterministic policy engines.

This release formalizes ButterClaw's expansion into Fleet-Scale Awareness. The system introduces a persistent cross-session trust graph, multi-agent collusion detection, and a unified memory substrate that synthesizes spatial kinetic telemetry into semantic threat attractors. We present the ButterClaw ecosystem as both a practical engineering package and a conceptual ontology for the next generation of resilient AI safety infrastructure.