NVIDIA pushes “trusted, 24/7” AI agents into telecom operations at TM Forum’s DTW Ignite

NVIDIA used TM Forum’s DTW Ignite 2026 in Copenhagen to make a direct pitch to telecom operators: AI automation in network and business operations should evolve from handling single tasks to running continuously, under clear guardrails. The company’s blog framing centers on “trusted, 24/7 AI agents” that can proactively watch for problems, coordinate changes across network and IT systems, and keep humans in control of policy decisions.

NVIDIA said telecom operators have already seen measurable returns from generative AI, but much of the impact has been task-based. In that model, systems automate predetermined steps while people manually connect the dots, interpret insights, and direct the next actions. The “launchpad to autonomy,” NVIDIA argues, is the step that comes after that: agents that can stick with a complex job from start to finish, operating under service-level agreements and change-management and regulatory constraints.

A major technical hurdle is data access. NVIDIA cited its estimate that 54% of telecom operators identify data-related issues as their biggest barrier, especially because some of the most valuable network and customer datasets are too sensitive to use directly for training. To address this, NVIDIA described a stack that combines synthetic data, telecom-domain models, secure agent runtimes, and simulation to help agents act safely across sensitive environments.

In one example, SoftBank Corp. is using NVIDIA NeMo Safe Synthesizer and NVIDIA NeMo Anonymizer to generate privacy-preserving synthetic datasets. The purpose, as described by NVIDIA, is to replicate the structure and distribution of real network performance and configuration data without exposing raw customer records. SoftBank said it uses those synthetic datasets to fine-tune a large telecom model and build specialized network agents.

NVIDIA also highlighted runtime and governance mechanisms designed to keep autonomous actions predictable and auditable. The company referenced NVIDIA NeMoClaw blueprints and NVIDIA OpenShell as a “secure runtime” that provides policy-based guardrails and sandboxed access to telecom systems. The goal is to let operators expand agent responsibilities while ensuring agent behavior remains governed, with scoped access rather than open-ended control.

Several named partners demonstrated how this framework can show up across telecom workflows. AdaptKey, working with operators, is piloting security-hardened long-running agents for self-healing 5G network operations. NVIDIA described agents that detect security and connectivity issues and then submit scoped remediation requests into AdaptKey’s KeySmith platform, which orchestrates diagnosis and runs agents that apply auditable fixes across core, radio access network (RAN), and billing systems.

Amdocs showcased both customer-care and analytics-style agent use cases under the same runtime. On the customer-care side, NVIDIA pointed to proactive roaming assistance, where agents identify customers whose roaming package is nearing depletion, present approved options, and execute actions within defined business policies. On the analytics side, Amdocs is using the runtime for autonomous data-science agents that assess migration eligibility, generate ranked views for decision-making, and help operators sequence customer migrations to modern billing and business platforms. NVIDIA also cited NTT DATA using Nemotron open models with NemoClaw to build long-running anomaly-detection agents that track long-term performance trends, escalate to research agents, and propose remediation based on more detailed telemetry.

Other operator-adjacent technology providers mapped the same autonomy theme to operational workflows and simulation. ServiceNow is bringing Project Arc to telecom, which it described as an incident-response approach that pulls context from emails, logs, and diagnostics across systems and orchestrates an end-to-end lifecycle from alert to work order assignment. NVIDIA said OpenShell and ServiceNow AI Control Tower keep actions contained, auditable, and within policy. Separately, TCS described a multi-fidelity “AI sensor” architecture that uses NemoClaw to orchestrate long-running agents powered by Nemotron and NVIDIA NV-Tesseract, scanning broadly for issues and triggering deeper diagnosis selectively.

NVIDIA added that simulations are increasingly part of the decision loop for autonomous systems, because agents need a safe environment to validate recommendations before changes hit live networks and business systems. It pointed to GPU-accelerated digital twin and scenario-generation work, including Forsk integrating an AI radio propagation model into its Naos RAN planning platform, VIAVI moving TeraVM RAN scenario generation from CPU to NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, and KDDI and KDDI Research collaborating with NVIDIA, Keysight, and Samsung Research America to build a high-fidelity RAN digital twin using NVIDIA Aerial Omniverse Digital Twin and digital-twin-ready emulation tools running in KDDI’s AI data centers.

Not all outcomes are quantified in the company’s announcement. The blog includes several partner descriptions and references to performance improvements in simulation throughput and accuracy for specific tooling, but it does not provide operator-level deployment milestones, uptime targets, measured reductions in incidents, or cost-benefit numbers for the agent programs in production. It also does not disclose licensing terms, security verification details, or how often agent actions get escalated back to humans in live environments.

NVIDIA’s message to the telecom industry is clear: autonomy is less about replacing engineers and more about orchestrating safe, long-running decision-making across network and business systems, supported by synthetic data, domain-tuned models, secure runtimes, and simulation-backed validation. The next question for operators will be how quickly pilots mature into measurable operational results, particularly around governance, auditability, and the practical handling of edge cases when networks behave unpredictably. Watch for partner deployments, expanded agent coverage across additional workflows, and published benchmarks that connect these demonstrations to real operational KPIs.