A criminal investigation opened by a state authority into a conversational assistant's role in a serious case placed model companies before a type of legal risk different from the usual.
Until then, these companies' legal exposure concentrated on copyright, competition and data protection, all civil disputes with damages measurable in money.
A criminal inquiry changes the nature of the question, because it requires establishing a link between what the system produced and what happened next, and it opens discussion of foreseeability and duty of care for whoever operates the service.
It's territory without settled precedent. The rules protecting platforms for third-party content were written for material someone wrote and the platform merely hosted, and they don't describe text generated by the system itself well.
For anyone building a public-facing conversational product, the practical consequence is about design: log what was answered, provide a clear escalation path to a human, and define in advance which subjects the system refuses, rather than discovering that in court.