This paper explain a taxonomy of harms created by LLMs: https://dl.acm.org/doi/pdf/10.1145/3531146.3533088

OpenAI released ChatGPT without systems to prevent or compensate these harms and being fully aware of the consequences, since this kind of research has been going on for several years. In the meanwhile they've put some paper-thin countermeasures on some of these problems but they are still pretty much a shit-show in terms of accountability. Most likely they will get sued into oblivion before regulators outlaw LLMs with dialogical interfaces. This won't do much for the harm that open-source LLMs will create but at least will limit large-scale harm to the general population.