Liability should always fall back to the whoever is running the bot.

That said, a public-facing AI with access to internal data is itself a security risk. People bully and cajole AI's into disclosing private information and sometimes it is laughably easy to do so.