Agent Capability Control (ACC): Capability-Based Security for Autonomous AI Agents

This paper introduces Agent Capability Control (ACC), a capability-based authorization framework for autonomous AI agent ecosystems. ACC ensures that skills declare required permissions, agents declare granted capabilities, sub-agents receive attenuated permissions through monotonic reduction, and every authorization decision is auditable. Key contributions include capability tokens with cryptographic binding, hierarchical attenuation ensuring sub-agents never exceed parent permissions, and integration with FDAA, GAM, and DCT.