LLMs improve network control plan safety on live 5G systems
Measuring Bounded Autonomy for an LLM Decomposition Planner on a Live 5G-Core Testbed
Networking and Internet Architecture
Summary
Large language models (LLMs) are being used to help manage complex 5G networks, but sometimes their plans include actions that aren’t allowed. The authors tested a system where the LLM’s output is restricted by rules and checked continuously to ensure it only suggests safe, executable steps. They found that without these limits, mistakes were common, but with the rules enforced, the LLM produced only valid plans. This means LLMs can be trusted more for network tasks if guided carefully and their decisions distilled into simpler rules.
What this means in practice
- •For network operators: Ensure safe automated control of 5G network procedures by constraining AI planner outputs to executable actions and verifying them live.
- •For telecommunications developers: Build AI-assisted control systems that reliably decompose complex network tasks without violating operational constraints using schema-based bounds.
Authors
Sanjeet Raj Pandey, Axel Küpper
Abstract
Large language models (LLMs) are entering network-management control loops, yet their plans can fall outside the executable action space. Whether hard deterministic bounds around such a planner have real effect, or merely restate a definition, is an empirical question. We measure it on a live 5G-core testbed, where an LLM planner decides when to decompose control-plane procedures into stateless handlers inside a bounded loop: a JSON schema constrains the output, a projection-repair validator re-checks it, a deterministic fallback guarantees liveness, and every decision is persisted. Three measurements anchor the paper. First, the two tested models, run unbounded with the allowed action sets stated in the prompt, still violated intent-handler rules in 13 of 44 standard calls. Every violation broke an intent-handler combination rule that the deployed flat schema cannot encode (schema enforcement as deployed caught 0/13); a union schema encoding the rule removes the class by construction. Second, with the bounds in place, all 33 decoded live plans were feasible without repair, and a matched-condition probe put the in-distribution unbounded rate at one violation in 20 windows. Third, exploratory: on the design set the planner matched an author-specified selection reference that a threshold policy and three a priori codings each missed. Transfer to held-out scenarios remained partial (5-6 of 10), and the planner's two-clause distillate could replace the runtime planner itself. Together, the results suggest elicitation and distillation, not runtime control, as the LLM's role.