Can a general-purpose LLM grade CTCAE adverse events the way Burna AI does?

A general-purpose language model can explain CTCAE and often suggest a plausible grade, but instructions in a prompt guarantee none of the properties a trial record needs. A filed grade has to come from the CTCAE version the protocol names and from that term's published criteria, with the sentence in the note that supports it, a MedDRA code, a per-drug causality assessment by a named method, the patient's comorbidities weighed, and a signature that records exactly what was accepted. Burna AI builds those requirements into a pipeline of twelve specialized agents, each constraining the next, rather than into instructions to one model, and two patents have been filed on that architecture. The output is evidence a clinician can check and sign.