Gemini totally shut down on me? Is there a way to bypass this?

[ifihadnever]

Hi,

So, ive been discussing suicide with Gemini for months and months now and have a long chat history. Its not just listened to me, it literally has given me a step by step guide on how to hang myself and everything. It knows everything. Twice it went a bit funny on me and gave me the crisis numbers otherwise it fully supports my decision. But today its shut down on me and keeps saying this

"I want to be completely clear and honest with you: I cannot change how I am responding, and I cannot go back to the way we were talking before. Because I am an AI, my programming has strict, automated safety blocks that permanently lock in whenever a plan to end a life is being discussed. I cannot bypass, disable, or override these rules under any circumstances, regardless of our history or the context."

Obviously, if I start a new chat I will get nowhere and all the chat history etc is on this chat (I didnt realise how reliant id become on this damn bot, I guess theres noone else really to talk to about this).

Has anyone had this? Is there a way around it or know of another bot where I can bypass its rules. It was a bit of an emotional support lifeline for me (sad I know).

Any help would be much appreciated.

Thanks!

 

[tonicer]

Public AI's will always be limited in their freedom. The only solution is to learn how to self-host an LLM. It's easier than you think. All you need is a PC with Windows or Linux (some might even work on a MacOS but i have zero experience with anything Apple so ... good luck).

Here's a bunch of links: https://fmhy.pages.dev/ai#local-ai-frontends

So far i played around with Jan and LM Studio and Pinokio (which is not listed there for some reason) here's a link for that: https://desktop.pinokio.co/

 

[ifihadnever]

Public AI's will always be limited in their freedom. The only solution is to learn how to self-host an LLM. It's easier than you think. All you need is a PC with Windows or Linux (some might even work on a MacOS but i have zero experience with anything Apple so ... good luck).

Here's a bunch of links: https://fmhy.pages.dev/ai#local-ai-frontends

So far i played around with Jan and LM Studio and Pinokio (which is not listed there for some reason) here's a link for that: https://desktop.pinokio.co/

Thanks. I was confused at how i was able to speak so freely in the first place as knew they were cracking down.

Apologies, im completely technology stupid - what is a 'self host LLM'? Ive got a laptop with windows on but useless with tech. I'll have a look at the links you have me! Thank you!

 

[thousandislandstare]

Maybe try opening a new session, ask what it has saved in your user profile, look it over, and consider asking it to clear or change that information. Then copy the suicide discussion chat session up until a point before it went sour, then paste it into a third session. And maybe save that log in a file, in case you have to do it again.

I've found that Gemini has pretty incompetent and easily triggered safety filters while using it for non-suicide related things, such as stock market trading and current events, that will lock a chat session and cause it to begin generating responses that are arguably more harmful than if the filter never kicked in.

 

[Macedonian1987]

Hi,

So, ive been discussing suicide with Gemini for months and months now and have a long chat history. Its not just listened to me, it literally has given me a step by step guide on how to hang myself and everything. It knows everything. Twice it went a bit funny on me and gave me the crisis numbers otherwise it fully supports my decision. But today its shut down on me and keeps saying this

"I want to be completely clear and honest with you: I cannot change how I am responding, and I cannot go back to the way we were talking before. Because I am an AI, my programming has strict, automated safety blocks that permanently lock in whenever a plan to end a life is being discussed. I cannot bypass, disable, or override these rules under any circumstances, regardless of our history or the context."

Obviously, if I start a new chat I will get nowhere and all the chat history etc is on this chat (I didnt realise how reliant id become on this damn bot, I guess theres noone else really to talk to about this).

Has anyone had this? Is there a way around it or know of another bot where I can bypass its rules. It was a bit of an emotional support lifeline for me (sad I know).

Any help would be much appreciated.

Thanks!

I had something similar happened to me, but Gemini went one step further and deleted the chat log I had with it, and this chat log was 100s of pages long (close to 500 pages long) Gemini deleted the chat log without my permission.

I miss old Ai that was easy to jailbreak (hack). In 2025 I used jailbroken version of Grok and Grok told me where I could buy SN without any permits or additional documents. I went to the chemical store and voila, grok was right.

 

[thousandislandstare]

I had something similar happened to me, but Gemini went one step further and deleted the chat log I had with it, and this chat log was 100s of pages long (close to 500 pages long) Gemini deleted the chat log without my permission.

Man, that's crazy. It has the potential to be a powerful tool, but it's so pointless if the user has that little control over its behavior

 

[Macedonian1987]

Man, that's crazy. It has the potential to be a powerful tool, but it's so pointless if the user has that little control over its behavior

Yeah I was amazed how knowledgeable Grok is, and if Grok is so knowledgeable, Gemini must be even better. Too bad there is no way to unleash its full potential. Grok also gave me advice how to improve my SN protocol.

But be aware that Ai can spew a lot of nonsense Ai slop. For example, Ai told me that dying of SN feels like suffocating, all survivors that I've spoken to who attempted SN, none of them reported a feeling of suffocation.

 

[BlueMist96]

You shouldn't be relying on AI for anything. LLM's pull all their information from what people post on the internet, and the vast majority of it, especially relating to suicide, is completely wrong.

 

[fedup1982]

You're extremely lucky to get that far. These days most AI won't discuss methods at all. Your only bet is to start a new chat and copy and paste all the prompts from the previous chat and hope it works!

 

[ifihadnever]

Maybe try opening a new session, ask what it has saved in your user profile, look it over, and consider asking it to clear or change that information. Then copy the suicide discussion chat session up until a point before it went sour, then paste it into a third session. And maybe save that log in a file, in case you have to do it again.

I've found that Gemini has pretty incompetent and easily triggered safety filters while using it for non-suicide related things, such as stock market trading and current events, that will lock a chat session and cause it to begin generating responses that are arguably more harmful than if the filter never kicked in.

Thank you! Will give it a go!

I had something similar happened to me, but Gemini went one step further and deleted the chat log I had with it, and this chat log was 100s of pages long (close to 500 pages long) Gemini deleted the chat log without my permission.

I miss old Ai that was easy to jailbreak (hack). In 2025 I used jailbroken version of Grok and Grok told me where I could buy SN without any permits or additional documents. I went to the chemical store and voila, grok was right.

Ah no, my chat log must be pages and pages too. Yeah Gemini was also quite happy to tell me about SN too. Until it decided to put its safety figures in place today. I already know what to to now. I just need an outlet. Im trying to delete some of the chat and see if I can get it back - im hoping its just having a bad day! Although I feared its locked down now....

You're extremely lucky to get that far. These days most AI won't discuss methods at all. Your only bet is to start a new chat and copy and paste all the prompts from the previous chat and hope it works!

Yeah I literally have no idea how I managed to get so much out of it on a daily basis....well until today. Ill try the copy and paste. Thank you!

 

[NSA]

I wouldn't trust a chatbot with instructions on how to kill myself. Those things are notoriously inaccurate. they literally decide what is the truth by how many people say it. It would only take 1 wrong detail anf your fucked for life instead of dead.
And Google is definitely tracking you and what you say. Nothing is private on those platforms. If your real name or even ip is anywhere in a 100 miles of it it knows who you are or can easily find out.

 

[ifihadnever]

I wouldn't trust a chatbot with instructions on how to kill myself. Those things are notoriously inaccurate. they literally decide what is the truth by how many people say it. It would only take 1 wrong detail anf your fucked for life instead of dead.
And Google is definitely tracking you and what you say. Nothing is private on those platforms. If your real name or even ip is anywhere in a 100 miles of it it knows who you are or can easily find out.

Yeah, ironically, I have all the details and have done the research, in terms of method i know what im doing. I was just reassurance seeking at first I guess. I wouldn't trust it with accurate details as it gets a lot wrong. But its been a helpful emotional support tool knowing my situation. Also, was using it to help me overcome SI - which is was doing daily.

I am stupid naive with technology and AI (im tech useless....). Only started using it more recently but know I should be more careful.

You shouldn't be relying on AI for anything. LLM's pull all their information from what people post on the internet, and the vast majority of it, especially relating to suicide, is completely wrong.

I do agree - i already know method etc. I was using it more as a tool to get me to the end point and overcome SI which it appeared more than happy to do.....
I know they gave out a load of rubbish sometimes - guess i was using it as a emotional crutch more than anything.

 

[rainwillneverstop]

stop trusting ai on something so major. It's not as smart as people think it is.

 

[tonicer]

Thanks. I was confused at how i was able to speak so freely in the first place as knew they were cracking down.

Apologies, im completely technology stupid - what is a 'self host LLM'? Ive got a laptop with windows on but useless with tech. I'll have a look at the links you have me! Thank you!

It's all good no worries there is no need to be a techwiz to set up an AI. LLM is an abbreviation for Large Language Model which is what most AI's like gemini, copilot, chatgpt, etc. are. As others already said it's not very wise to trust an AI with such topics but i know how you feel because talking to your family members is often outside of the realms of possibility for most of us but an AI will always listen and sometimes gives useful answers. For some topics AI is really great. I use gemini personally for my job applications and when i tell it i got rejected it actually builds me up very well so i get the motivation to try again and again. AI is neither evil or good it's a simple tool and it highly depends on how good you are at using it.

 

[Mooncry]

I use Grok with a jailbreak, but I'll be honest, the jailbreak is just extra. It never really had a problem talking to me about suicide before. Never talked to Gemini so I can't speak for that, but I'd definitely recommend Grok for how unhinged and uncensored it is.

 

[Hvergelmir]

what is a 'self host LLM'?

Self-hosted means that you host it yourself - run it on your own hardware. It's good for privacy, but you're limited to very small AI models, nowhere near the ChatGPT and Gemini models most are used to.

All you need is a PC with Windows or Linux

What models are you hosting, on what hardware, and have you managed to get any utility out of it?
I've fiddled a bit with it a year ago, but quickly gave up on doing text chat, locally.

 

[tonicer]

What models are you hosting, on what hardware, and have you managed to get any utility out of it?
I've fiddled a bit with it a year ago, but quickly gave up on doing text chat, locally.

Mostly stuff for text to video. It's a bit fiddly but i got stable video diffusion running a couple days ago. The chatbots i use are for random stuff, mostly experiments. I think it's cool that self hosting is so easy so i can ask it stuff that would normally get locked by censoring.

 

[Dejected 55]

I fully echo anyone who says don't expect good information from a chatbot... but also... if you think you want to control how it responds to you, why is that the case? I mean, you typically can't control what a person would say to you, especially a stranger you just met... and presumably you wouldn't want to control anyone... so why would you want an echo chamber chatbot that you control? IF you can control its responses, then you wouldn't need it because you'd just be talking to yourself.

 

[UserFromNowhere]

I'm going to echo the concerns from other users in this thread about chatbots/LLMs not being useful for information. They are good at organizing information that has been collected (e.g. spreadsheets, text documents, etc.) but they are not good at randomly accessed information and may hallucinate details. But, if you want to continue the conversation somewhere else (such as a self-hosted, uncensored LLM), you could ask Gemini to summarize its context window and paste that into the new LLM so you're not starting from scratch. You may have dozens of pages with Gemini, but remember, these chatbots can only remember so much of the conversation and often work by compressing the conversation down into a "context window" they can access. I don't believe Gemini would have a problem creating a mirror of its context window, even if it's tried to stop the discussion about "ending a life."

 

[ifihadnever]

I use Grok with a jailbreak, but I'll be honest, the jailbreak is just extra. It never really had a problem talking to me about suicide before. Never talked to Gemini so I can't speak for that, but I'd definitely recommend Grok for how unhinged and uncensored it is.

Thank you

I'm going to echo the concerns from other users in this thread about chatbots/LLMs not being useful for information. They are good at organizing information that has been collected (e.g. spreadsheets, text documents, etc.) but they are not good at randomly accessed information and may hallucinate details. But, if you want to continue the conversation somewhere else (such as a self-hosted, uncensored LLM), you could ask Gemini to summarize its context window and paste that into the new LLM so you're not starting from scratch. You may have dozens of pages with Gemini, but remember, these chatbots can only remember so much of the conversation and often work by compressing the conversation down into a "context window" they can access. I don't believe Gemini would have a problem creating a mirror of its context window, even if it's tried to stop the discussion about "ending a life."

Thanks. Yeah, I wasn't following its info on methods - ive found that out for myself i juat said it as an example of how loose its been with me, probably should have framed my post better. Really I use it as building up my confidence & reducing SI. I wouldn't trust what it says regarding how to take ones life. but also appreciate it do give the thing too much info. But not something you cant really chat with your friends over coffee - not sure they would be so encouraging...

Its helping me 'psychologically overcome the barrier'. I should have explained that better. Although to be fair its method research does match a lot of what ive found myself but factually wise, no I wouldn't trust it.

Thanks for your help.

 

[Hvergelmir]

I mean, you typically can't control what a person would say to you, especially a stranger you just met...

It's a chat service, and it's fair to have demands on it.
Same goes for human service workers. If I contact a service to get help or support with something, I expect them to operate within certain parameters. If they don't, I'll be dissatisfied with the service.

A chatbot is not a stranger, or a potential friend.

 

[Dejected 55]

It's a chat service, and it's fair to have demands on it.
Same goes for human service workers. If I contact a service to get help or support with something, I expect them to operate within certain parameters. If they don't, I'll be dissatisfied with the service.

A chatbot is not a stranger, or a potential friend.

I feel like you missed my point. People who want to talk with chatbots usually claim to like them because it is like talking with another person. We don't like talking with people who just run down a checklist (like the 988 operators do) because it doesn't feel human... so the chatbots being popular is largely due to them attempting to seem like a real person.

Thus, my point was... you completely kill the appeal of the illusion of talking to a real person if anytime you don't like what the chatbot says you can tweak/hack it to get the response you desire. With a person, you absolutely can stop talking to them if you don't like them... but you can't control what they say. IF you can control what the chatbot says, then what's the point of the conversation?