Abstract
A self-improving multi-agent artificial intelligence (AI) system is proposed herein for impact-aware infrastructure configuration remediation across cloud, on-premises, network-fabric, container, and hybrid environments. The system builds a temporally weighted identity and dependency graph (TIDG) from telemetry and large language model (LLM)-extracted relationships, uses a graph neural network (GNN) to predict the operational blast radius of a proposed change, and routes each finding to one of four simulation tiers. Every tier emits a normalized per-principal good-actor impact vector. Independent security, reliability, cost, and compliance evaluators use that vector in an adversarial governance gate. Accepted changes advance through telemetry-gated observe, canary, progressive-ring, and lockdown phases, with automatic rollback on deviation. Deployment results and operator feedback become labeled training data for the dependency extractor, tier-selection policy, blast-radius predictor, and governance thresholds. The system also generates policy-as-code guardrails after successful remediation to reduce recurrence.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Jackson, Arica Schivonne; Mahadevaswamy, Sushma; Addepalli, Sree Gowri; and Yuan, Barry Qi, "SELF-IMPROVING MULTI-AI AGENT SYSTEM FOR IMPACT-AWARE INFRASTRUCTURE CONFIGURATION REMEDIATION", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12391