I mean, if you do and it hacks someone, you should be (and likely are) criminally liable
I think liability works for, e.g., an oil refinery that has an explosion every 30 years, but does liability work for an AI lab that is causing harm to society every week?