Frontier AI in Communications Infrastructure

Frontier AI in Communications Infrastructure examines what happens when the most capable classes of artificial intelligence are embedded directly inside the systems that route, secure, and manage critical communications. As operators adopt AI for traffic optimization, fraud and intrusion detection, and autonomous network management, the paper asks how accountability and oversight can keep pace with capability.

The core contribution is a function-based framework. Rather than regulating models by name or size, the paper proposes governing them according to the functions they perform inside a network — monitoring, decision-making, or direct control — and assigning oversight obligations proportionate to the consequences of failure in each function. This allows operators and regulators to reason about risk without waiting for a complete technical taxonomy of every system.

The analysis catalogs failure modes specific to communications settings: automation bias among operators, opaque model behavior during rare but high-impact events, data-poisoning and prompt-based manipulation of AI-driven controls, and the systemic risk created when many carriers rely on a small number of underlying models. It stresses that a single shared weakness can become a sector-wide vulnerability.

The brief offers practical guidance for operators, vendors, and policymakers: maintain human authority over consequential actions, require auditability and logging for AI decisions, test systems against adversarial conditions, and coordinate sector-wide standards before dependence becomes irreversible. It is published as a working paper on ResearchGate.