It's not that there aren't some risks to LLMs that produce illicit output, but the training data that made such output possible is openly available online already. As far as the risks of flooding the web with misinformation and propaganda, well...