Telecom Operators Face Control Questions as AI Agents Act on Networks

Telecom operators are deploying AI agents capable of making autonomous changes to live networks, creating urgent questions about governance, liability, and control. Major equipment vendors agree on one principle: operators must remain the ultimate authority over their infrastructure, even as artificial intelligence systems gain the ability to act without human intervention.

The shift from recommendations to actions

AI systems in telecom networks are evolving beyond advisory roles. Where earlier implementations suggested optimizations for human review, emerging agentic AI can execute changes directly—restarting network functions, modifying network slices, or adjusting configurations in response to conditions. This transition raises fundamental questions about who governs these decisions and who bears responsibility when automation fails.

Oguz Sunay, CTO of AI and Autonomous Networks at Nokia, described the target model as "glass-box governance," where agents operate within operator-defined boundaries rather than as independent authorities. "The operator's policy must govern the outcome," Sunay told Fierce Wireless. "AI can help identify the best action; the operator defines the conditions under which it can occur."

Guardrails and governance mechanisms

The vendors outlined specific control mechanisms operators should retain. Ericsson's Akhil Gokul, Head of Technology Strategy Office for the Americas, emphasized that where a model runs matters less than who sets the rules. Operators should define intent and policies, govern data access, control agent permissions, retain authority over network changes, and maintain visibility through audit trails.

Arjun Nanjundappa, staff engineer at Samsung Networks, argued operators need not own every AI stack component but should control the reasoning loop, tool integration, context management, and guardrails. He pointed to bounded actions as a practical safeguard—limiting an agent to specific network slices or functions prevents a faulty remediation from cascading across infrastructure.

Why it matters

The liability question remains unresolved. When an AI-driven change causes a service outage, responsibility depends on deployment design, commercial agreements, agent authority, and applicable law. Without clear audit trails showing what actions were taken and under what authority, operators face both operational and legal exposure. As networks become more autonomous, the governance frameworks established now will determine whether operators maintain strategic control or cede decision-making authority to vendors and hyperscalers.

The hyperscaler tension

Hyperscalers present both opportunity and risk in this equation. They offer essential AI infrastructure, development tools, and model ecosystems, making them natural co-development partners. Yet vendors warned of lock-in shifting from hardware to the intelligence layer—the models, policies, and context that drive network decisions.

Sunay described this as an "intelligence gravity well," where the provider controlling the agent context layer could exert more influence over network behavior than whoever hosts the core workload. Both Nokia and Ericsson stressed that standards-based interfaces across organizations like O-RAN, 3GPP, TM Forum, and the AI-RAN Alliance are essential to keeping agentic automation portable and preventing one-way dependencies.

These details were first reported by Fierce Wireless.