Why OTR404
A short, honest read
When you talk to ChatGPT, Gemini, Claude, or Meta AI, you are talking on the record. The companies running those services keep the logs, train on them in some cases, and can be forced to hand them over. We built OTR404 because that is a real problem and it is going to get worse before it gets better. Below is the case for why, with the articles and videos that convinced us.
Your old chats already got preserved by court order
In May 2025 a federal judge in the New York Times copyright lawsuit ordered OpenAI to preserve every ChatGPT conversation, including ones users had deleted. By November the court made them produce a sample of 20 million chat logs to the plaintiffs. OpenAI fought the order and lost. Their own response to it says the data is now sitting in a legal hold accessible only to a small audited team. The point is, it exists. If you used ChatGPT between April and September 2025 and deleted something you would not want read by a stranger, it is still there.
More legal commentary on what this means for everyday users: National Law Review and Thurrott.
Sam Altman is telling you himself
Last summer Sam Altman went on Theo Von's podcast and basically said the quiet part out loud: there is no legal privilege for what you tell ChatGPT. If you tell a therapist, lawyer, or doctor something, that conversation is protected. If you tell ChatGPT the same thing, OpenAI can be forced to hand it over. He called the situation "very screwed up" and asked for new laws. There are no new laws.
Police are already pulling chat logs as evidence
CNN ran a piece this morning, "How ChatGPT conversations became a treasure trove of evidence in criminal investigations". They walk through a double homicide case, the LA arson investigation, and a school shooting case, all of which used ChatGPT history. One cybersecurity expert calls AI chat logs a treasure trove because suspects type direct, incriminating questions assuming nobody will ever read them. They do read them.
Futurism wrote a similar piece specifically about people who used ChatGPT to ask legal questions and ended up with their own queries used against them in court.
Divorce lawyers figured this out fast
After the Heppner ruling, AI chat history is officially discoverable in family court. If you used ChatGPT to vent about your marriage, ask about hiding assets, or work through whether to leave, the other side can subpoena it. Divorce attorneys in New York are already doing it. Florida firms walk you through the process on their own websites now.
Your partner does not need a subpoena
The way most people get caught is way simpler than that. Someone borrows your laptop. The ChatGPT sidebar is right there with conversation titles like "relationship issues" or "how do I tell her." Slate ran a first-person piece in April from a woman who saw her boyfriend's ChatGPT history and the relationship ended that night.
People treat ChatGPT like a journal. It is not a journal. It is a sidebar full of titles, on a device that other people sometimes touch, backed by a server that has been compelled to retain everything.
Meta is now using AI chats to target ads
On December 16, 2025, Meta started using your conversations with Meta AI across Facebook, Instagram, and WhatsApp to personalize ads. You cannot opt out. The EU, UK, and South Korea are exempt because their privacy laws made Meta back off there. The US is not exempt.
EPIC and a coalition of consumer groups asked the FTC to stop it. Nothing happened. Proton has a plain explainer of what is and is not under your control. The short version: not much. So if you ask Meta AI about a breakup, a medication, a job change, anything, expect the ad system to act on it.
Gizmodo's coverage of the policy change spells out what Meta will and will not exclude. They claim sensitive topics like religion and health are off limits, but to classify a conversation as sensitive they still have to read it.
What we built instead
OTR404 is the same kind of AI you are already using. Chat, image, video, audio, code. The difference is what we do not do. We do not store conversations. They live in memory for the length of your session and then they are gone. We route requests through zero-retention endpoints so the model providers do not keep them either. There is no chat history page because there is no chat history.
We do not run ads. We do not sell data. We do not train on your conversations. There is no logging pipeline that could leak, no archive that could be indexed, no chat sidebar that could be borrowed. Your conversations live in memory for the length of your session and then they are gone — not in a database under our control, not at any provider in the request path. The only way to guarantee something stays private is to never collect it in the first place.
To be clear about what privacy means here: it is privacy for you, not cover for harm to others. We have an Acceptable Use Policy that prohibits using the Service for things like CSAM, fraud, harassment, malware development, or attacks on infrastructure, and we cooperate with lawful investigations into those uses within the limits of what we hold. What we hold is small on purpose.
That is the whole pitch. If it sounds useful, the private beta is open on the homepage.