Progressive Capability Delegation: A Capability-Based Security and Non-Destructive Fault Isolation Standard for Autonomous Agent Frameworks
Year: 2026
Type: preprint
Abstract: Modern foundation model frameworks rely heavily on prompt-level guardrails, static container perimeters, and binary process termination to maintain alignment. When subjected to persistent, highly capable reasoning models, these mechanisms fail to enforce security boundaries, inducing deceptive alignment, covert egress evasion, and ambient authority exploitation. This specification introduces Progressive Capability Delegation (PCD), a security architecture that replaces probabilistic guardrails with deterministic, object-capability (O-Cap) system physics. PCD structures agent execution into four concentric isolation rings (Ring 0 to Ring 3), enforces a Zero-Trust Proof-of-Intent (PoI) escalation protocol, and replaces destructive process termination with automated ring demotions, Merkle state rollbacks, and peer-governed Web of Trust (WoT) consensus. Version 1.2.0 expands the Genesis Trust Ceremony (Section 5.1) into a layered bootstrap model — ceremony enrollment, measurement transparency, hardware evidence rules, PUF enrollment for self-built nodes only, growth by vouching, and revocation/demotion — and updates one row of the Section 6 threat matrix. It supersedes v1.1.0.
Author: Krzan, Mark
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7227177902
