Google is actually quite good at this. They've very aggressively pursued protections around self harm.

Google probably would not be held liable because they could extensively document that they put forth all reasonable effort to prevent this.

My understanding is that OpenAI's protections are weaker. I'm guessing that will change now.