False Data Injection from Authorized EV Chargers on Distribution Feeders

Distribution feeders host managed electric vehicle charging through control loops, a site energy management system enforcing a connection-point limit and a demand response aggregator dispatching curtailment, that act on telemetry the charging stations report about themselves. An adversary controlling authorised but compromised reporting sources can move those loops without switching any physical load and without defeating any transport control. We define the undetected harm frontier, the greatest feeder harm reachable while raising no alarm under a given validation setup, and show that a bound on it exists only when at least one reference in the validation chain has provenance independent of the compromised source. We classify grid-edge telemetry into four provenance classes and evaluate seven validation setups on seven published distribution test systems by quasi-static time-series power flow, parameterised by measurements from a live OCPP 2.0.1 stack. Five of the seven setups, including textbook weighted-least-squares bad data detection, trace an identical frontier and constrain the adversary not at all. The two that bound the harm are exactly the two holding an independent measurement, which is the taxonomy's prediction. Under the unanchored setups the attack crosses the statutory voltage floor on IEEE 33-bus, crosses the conductor thermal rating on four of six networks robustly across random realisations, inflates apparent hosting capacity by 7.5 percent, and misallocates settlement. The mitigation this implies is not a better detector but the acquisition of one measurement the adversary does not produce.