How easily is to get reported/detained when talking with AI chatbots about suicidality?

[noname223]

I got the message that my questions were inappropriate for the chatbot in the past. But I never was extremely explicit about my suicidality. People on here use jailbroken AI chatbots to plan their suicide. I think that's way too risky for such a high stake decision.

But are there red lines which shouldn't be crossed when talking about suicide with a normal LLM? How easily will you get a visit by the police? There was an incident where the chatbot was very concerned (probably the company was scared about liability) and said something like. You now have to be very honest are you currently at risk. Obviously I replied no I am fine don't worry. But I am not sure how safe it is to talk with them about my mental health issues. I haven't been very open thus far. Because I could imagine they called the cops on me. This is also why I love SaSu so much. You don't have to worry about that.

 

[skyisblue]

If you do wish to vent to AI, you should be more vague because after the attack at FSU and numerous suicides, AI companies have been releasing way more safety features in their new models.

I'm not completely sure about content moderation, but I don't think your AI chatbot will report you unless you are really going crazy in your prompts.

But I am not sure how safe it is to talk with them about my mental health issues. I haven't been very open thus far. Because I could imagine they called the cops on me.

You shouldn't worry too much. Most chatbots just spam those silly "help is available" numbers. If your prompts were too revealing, then i think the most that happens is that your chat gets flagged and then your account is restricted while a human looks over your chat logs. I seriously doubt you would have emergency services called, just an account ban at worse.

 

[overthrone]

i could be wrong, but there's almost no regulation on it. There's been multiple cases of people ctb with advice from chatbots. I dont recommend it though, most bots have strict safety features and likely won't tell you anything useful. Plus, yknow, the environment.

 

[Aflame5926]

i dont think we are there yet which goverment will do a health check based on what you searched. i got locked out of claud 1 time but it just stopped all questions regarding suicide. so i just stopped using it and used google again.

but i do see this in the feature for a concern through

 

[cipher]

i usually get around it by framing prompts as if i were doing research for a novel, like doing character studies or explorations without explicitly identifying myself as the subject (even though the "character" is based on me).

however you would still have to be conscientious of whatever guardrails your llm of choice has. they won't really give specifics in regards to methods or how to prepare but you can probably skirt the line a bit with this approach by saying the chat is about "fictional characters" whenever it starts to show concern. obviously your mileage may vary, i personally haven't seen the "help is available" pop-up ever since i started chatting to it like this

 

[RadiantNumber]

I don't think LLMs operators would do that but I am just a random person