Suspected AI-Assisted Threat Communications Against Critical Infrastructure: A Protocol for an Evidence-Graded Incident Register and Pilot Case Reconstructions

Artificial intelligence (AI), including generative text and synthetic voice systems, may lower the cost of producing deceptive communications, impersonating trusted individuals, and distributing threats at scale. These capabilities create potential risks for critical infrastructure, including aviation, public broadcasting, education, healthcare, energy, and public administration. However, the presence of AI in malicious communications must not be inferred merely from automation, unusual wording, synthetic-sounding audio, or the disruption caused by a threat. This protocol establishes an evidence-graded incident register for suspected AI-assisted threat communications affecting critical infrastructure and essential public services. It defines five evidence tiers (T0–T4), distinguishes “None” from “Unknown,” specifies source-weighting and deduplication procedures, and establishes a reproducible coding framework. It also proposes hypotheses concerning the frequency, functional contribution, distribution, and institutional effects of AI-assisted threat communications. Each hypothesis includes an operational outcome, relevant confounders, and a falsification condition. Pilot reconstructions illustrate the distinction between documented AI use, historical automated threats, and recent incidents for which AI involvement or common attribution has not been established. Pilot case reconstructions and pre-specified operational review thresholds are included to evaluate the usability and limitations of the protocol. No coding benchmark has yet been conducted; benchmark results will be reported separately once available. The study does not presume that a coordinated campaign exists, that AI caused the recent increase in reported threats, or that all incidents without discovered explosives were deliberate hoaxes. Its purpose is to establish a transparent method for testing these propositions.