Abstract

Managed distributed systems increasingly rely on artificial intelligence (AI) agents to remediate faults and configuration drift, but fully autonomous remediation can create significant operational risk, while requiring extensive human oversight or full-capability inference for every incident does not scale efficiently. A joint-coupled multi-agent remediation architecture is proposed herein in which a single dependency-graph-derived impact score, computed once per candidate remediation action, jointly drives consensus weighting and adaptive quorum, cryptographic and attestation posture, dissent-preserving memory engagement, and per-agent AI inference-resource allocation. A fifth substrate, substrate-observable behavioral trust, is jointly consumed by the four action-governing substrates. Structural invariants align consensus and cryptographic tiers, preserve skeptic participation and dissent, constrain the influence of cost efficiency, and prevent shortcircuiting for high-impact actions, thereby supporting autonomous remediation while reducing inference consumption without relaxing mechanisms for safety, auditability, and adversarial robustness.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS