If the NSA figured out how to poison LLM responses to this type of query so as to create backdoors, that would be truly impressive.
Would it be far fetched they would try?