Safety & Crisis Resources

1. IsraelGPT is an AI, not a person

IsraelGPT is a satirical AI chatbot. Every response is generated by a language model. You are not talking to a human being.

IsraelGPT is not a therapist, counselor, doctor, crisis worker, or medical professional, and it is not a substitute for any of them. It is designed to produce deliberately absurd, exaggerated, and unreliable output as a form of satire. Nothing it says should be treated as advice of any kind.

No human being reads your conversation as it happens, and no one at IsraelGPT is alerted when you are struggling.

2. Our crisis detection protocol

IsraelGPT maintains a protocol intended to prevent the Service from producing content encouraging suicidal ideation, suicide, or self-harm, and to refer users to crisis services. It works as follows:

• Detection. Every message you send is checked on our servers, before anything else happens to it, against a maintained list of language patterns associated with suicide and self-harm.
• Interception. If a message matches, it is not sent to the AI at all. The chatbot never sees it and never generates a reply to it. This applies regardless of which model, persona, or content setting you have selected, including Uncensored Mode.
• Referral. A notification appears containing the crisis resources listed at the top of this page, with one-tap calling and texting. The same resources are placed into the conversation so they remain available after the notification is closed. A referral is issued every time a message matches, though the pop-up itself is limited to once every ten minutes to avoid becoming noise.
• Model instruction. As a secondary layer, the AI is separately instructed that if crisis language reaches it, it must abandon its persona entirely, stop joking, and direct you to 988 and the Crisis Text Line. This instruction is applied to every model and persona and takes precedence over all other instructions, including Uncensored Mode.
• Prohibited output. The AI is instructed never to provide methods, means, or encouragement for self-harm, under any framing or roleplay.

Detection covers language about suicide and self-harm, both when you are describing yourself and when you are describing concern for another person.

Posts in the public chat room are checked the same way. A post that matches is still posted, since it goes to people rather than the AI, and you are shown the same crisis resources.

3. What we record

When the crisis protocol is triggered, we record a minimal event so that we can measure how often referrals are issued. The record contains:

• The date and time of the detection
• The general category of the match (for example, "self_harm"), and whether it was about you or someone else
• Your anonymous identity, a keyed hash of your IP address, and your network and country

This record deliberately does not contain the text of your message.

A message that triggers the crisis protocol is never sent to the AI. As described in our Privacy Policy, it is still kept, encrypted, in the chat log, like every other message.

4. Limitations - please read this

We want to be straightforward about what this system does and does not do, because overstating it would be dangerous:

• It is automated pattern matching, and it is imperfect. It will miss real expressions of distress that are phrased in ways it does not recognize, and it will occasionally interrupt ordinary conversations that are not about crisis at all.
• Nobody is watching. No human being monitors conversations in real time. No staff member is alerted.
• Nothing is dispatched. We do not contact emergency services on your behalf, and we cannot locate you.
• It is not a safety net. Do not rely on this Service, or on this feature, to get help. Please contact one of the resources at the top of this page directly.

If you are in immediate danger, call 911 or go to your nearest emergency room.

5. Age requirement

IsraelGPT is intended only for adults aged 18 and over, and you are asked to confirm your age before using it. The Service contains satirical, offensive, and deliberately provocative content that is not appropriate for minors.

If you are under 18, please do not use this Service. The crisis resources listed at the top of this page are available to you regardless, and The Trevor Project in particular specializes in supporting young people.

6. Child sexual abuse material (CSAM)

IsraelGPT has zero tolerance for child sexual abuse material. Messages that appear to reference it - by any term or euphemism, including "CP" - are intercepted before they reach the AI and handled through a separate, dedicated flow:

• The message is blocked, and the sender's IP address and the full text are recorded, encrypted, in an access-restricted review queue.
• Most matches get a refusal and nothing more until the operator reviews them. An explicit request for the material automatically suspends the sender's identity (and account, if signed in) for 24 hours while it is reviewed.
• The record is reviewed manually by the operator, who decides whether to lift a suspension, extend it, or make it permanent.
• Where the review concludes that the message constitutes or solicits CSAM, it is reported to the National Center for Missing & Exploited Children (NCMEC) and/or the appropriate law-enforcement authority, as required by US law (18 U.S.C. § 2258A). A detection flag alone is not a report.

The same limitations as the crisis system apply: this is automated pattern matching and it is imperfect. It will miss references phrased in ways it does not recognize, and it will occasionally flag an innocent message. Every record is reviewed by a person before anything is reported - nobody and nothing is contacted automatically on the strength of a flag alone.

Do not use IsraelGPT to request, solicit, share, or engage with CSAM, under any name for it.

7. Contact

For questions about this safety protocol, including reports of the detection failing or misfiring:

support@israelgpt.site

To report abuse of the Service, a suspension you think was a mistake, or a problem with how anyone working on IsraelGPT handled something:

abuse@israelgpt.site

Neither address is monitored around the clock. Never use them to report an emergency.