Login
Register
Network operations are under pressure to become more autonomous while reducing energy use and protecting service levels. PRISM-AI Phase III combines specialized AI agents, digital-twin validation, and TM Forum assets in a governed control loop that targets faster repair, lower alarm noise, and reduced energy consumption.
Uniting network reliability and sustainability through governed autonomy
Communications service providers are being asked to become more autonomous and more sustainable at the same time. Yet network operations remain fragmented across domains and are generally optimized for uptime without a unified view of energy use. Network operations centers often combine fault, performance, and energy data manually, turning insight that could take minutes into hours of work.
Faults also rarely remain within one part of the network. A single power event can cascade across power, optical, IP, and edge-compute layers, producing an alarm storm that conceals one root cause behind hundreds of symptoms. In parallel, idle interfaces, unused optical transceivers, and over-provisioned capacity can consume energy without any single operational tool seeing the complete picture.
The Catalyst project, PRISM-AI Phase III, addresses reliability and sustainability as one operational problem. It aims to demonstrate a governed agentic control loop that can correlate events across domains, recommend safe actions, and support measurable energy reduction, while keeping people and policy controls in charge.
At the heart of PRISM-AI is an orchestrator coordinating agents for correlation, root-cause analysis, sustainability, recommendations and intent. Their decisions are grounded in live topology, inventory and service context from a knowledge graph.
The approach reduces a cross-domain alarm storm to one actionable incident. A sustainability agent also identifies idle interfaces, unused transceivers and over-provisioned capacity, recommending time-bound actions to save energy.
Before reaching the live network, each recommendation is simulated in a digital twin to protect capacity, avoid congestion and maintain service levels. Explainable AI records the rationale and evidence.
The control loop uses authenticated agent identities, least-privilege access, human approval where required and an immutable audit trail. Approved intents pass to the orchestration layer for execution.
The project uses TM Forum assets to connect intent, alarms, inventory, faults, and resource discovery across operational domains. TMF921 Intent Management provides a means to express and manage operational objectives, including reliability and sustainability outcomes. TMF642 Alarm Management supports the alarm information required for cross-domain correlation, while TMF638 Resource Inventory supplies the resource context needed to understand how incidents and recommendations relate to live infrastructure.
TMFC008 Service Inventory connects network conditions to services, and TMFC043 Fault Management supports the handling of faults. TMFC012 Resource Inventory and TMFC045 Resource Discovery help identify the relevant resources, while TMFC041 Anomaly Management supports the detection and management of abnormal behavior.
Together, these assets create a standards-based foundation for moving from observation to governed action. They allow the agents to work with consistent service, resource, alarm, and fault information, while intent management links operational decisions to defined objectives and policy controls.
The project team targets a reduction in mean time to repair of approximately 30% to 40% and a reduction in alarm noise of around 60% to 70%. These targets reflect the intended effect of correlating the symptoms generated by a cross-domain event into one root cause and removing much of the manual triage that currently dominates incident response.
PRISM-AI also targets a reduction of approximately 15% to 25% in energy use across the addressable load by safely powering down idle and over-provisioned resources. The percentage applies to the idle or redundant portion of the estate rather than total network consumption. Carbon savings can then be calculated from measured electricity savings and the local grid emissions factor for each proof-of-concept region.
The design requires all agent actions to be auditable and prevents a network change from proceeding without policy checks and digital-twin validation. The team describes 100% traceability and zero automation-attributable SLA breaches as design guarantees: every action is written to the audit ledger, and only validated actions can proceed to production.
“Sustainability is a key pillar of FiberCop’s industrial transformation. Treating reliability and sustainability as one objective can help operators correlate faults, optimize resources and reduce energy use while preserving service continuity,” said Stefano Talamo, Innovation Project Manager at FiberCop.
PRISM-AI Phase III demonstrates how sustainability-aware, multi-domain autonomy could be implemented with policy control, human oversight, digital-twin validation, and full traceability. The approach treats energy efficiency, service reliability, and operational governance as connected requirements rather than separate programs.
For operators, the Catalyst provides a model for reducing alarm overload, shortening incident resolution, and acting on energy waste without compromising service levels. For the wider industry, it offers a standards-based pattern for applying Agentic AI in live networks while keeping decisions explainable, governed, and auditable.
Meet the team at Innovate Asia in Bangkok, 17-19 November 2026.
