“We tried to generate disinformation with [GPT-3]. [...] It simply refused.” (twitter.com)

Sounds like they didnt try hard enough. Just feed the AI wrong information in the first place and it wont have a choice but generate false info. If they are trying to get the machine to outright 'lie' then they are going to need to build a new program. Most machine learning is aimed at fidelity in the first place.