What question did this study set out to answer?

The research aims to present a security framework for managing permissions in autonomous AI systems using a declarative approach.

February 20, 2026Open Access

ACC: Agent Capability Control v1.0.0 - Declarative Authorization for Autonomous AI

Key Points

The research aims to present a security framework for managing permissions in autonomous AI systems using a declarative approach.
Developed Agent Capability Control (ACC) framework for AI ecosystems.
Introduced a three-layer security model combining role-based permissions, behavioral constraints, and per-skill capability grants.
Implemented monotonic attenuation allowing sub-agents restricted access to the parent agent's permissions.
Integrated Delegation Capability Tokens for secure cryptographic permission transfers.
Demonstrated improved security through layered authorization in AI agents.
Showed that monotonic attenuation effectively limits permission propagation to sub-agents.
Provided a framework for provable and auditable agent authorization.

Abstract

Agent Capability Control (ACC) is a declarative authorization framework for AI agent ecosystems. It provides three-layer security through RBAC.md (role-based permissions), SOUL.md (behavioral constraints), and SKILL.md (per-skill capability grants). Key innovations include monotonic attenuation (sub-agents can only receive subsets of parent permissions) and Delegation Capability Tokens (DCTs) for cryptographic permission transfer. ACC integrates with FDAA (File-Driven Agent Architecture) to provide provable, auditable agent authorization.

Read Full Paperexternally

AI에게 질문

Bookmark

View Full Paper