https://ia801404.us.archive.org/31/items/gov.uscourts.nysd.6...
https://ia801404.us.archive.org/31/items/gov.uscourts.nysd.6...
It was doing this prior to being sued by the NYT and many others
OpenAI was collecting chats even when the user asked for deletion, i.e., the user did not want them saved
That's why a lawsuit could require OpenAi to issue a hold order, retain these chats for longer and produce them to another party in discovery
If OpenAI was not collecting these chats in the ordinary course of its business before being sued by the NYT and many others, then there would be no "deleted chats" for OpenAI to be compelled by court order to retain and produce to the plaintiffs
1. Or whatever period OpenAI decides on. It could change at any time for any reason. However OpenAI cannot change their retention policy to some shortened period after being sued. Google tried this a few years ago. It began destroying chats between employees after Google was on notice it was going to be sued by the US government and state AGs
This is an intelligence service's wet dream.
I don't think you even need to go as far as to Microsoft (who have earned zero points in the Privacy Protection league), just have a look at Altmans "I want to create a biometric database of every human" Orb/World-coin eye-scanning project: https://www.ft.com/content/0c5c2b8d-b185-40b6-9221-b80ee130b...
Im an age of automated backups and failovers, deleting can be really hard. Part of the answer could simply be that syncing a delete across all the redundancies (while ensuring those redundancies are reliable when a disaster happens and they need to recover or maintain uptime) may take days to weeks. Also the 30 days could be the limit, as oppose to the average or median time it takes.
I doubt it's that. Deletion is hard, but it's not "exactly 30 days" hard.
The most likely explanation is that OpenAI wants the ability to investigate abuse and / or publicly-made claims ("ChatGPT told my underage kid to <x>!" / "ChatGPT praised Hitler!"). If they delete chats right away, they're flying blind and you can claim anything you want.
Now, whether you should have a "delete" button that doesn't really delete stuff is another question.
Also I suspect there is a big gap between best practice and common practice. My guess is common practice is dysfunctional. I would also suspect there is no standard way, but there are established practices within different technology stacks that vary between performative, barely compliant and effective at scale.
In one case I saw there was a substantial manual effort to load snapshots into instances run the delete and then save new snapshots. This was over 10 years ago though and it was more of a "we just need to get this done" than a "what's the most elegant way to do this at scale"
When working on an e-commerce gig we would get "delete my data" requests from customers, which we're legally obliged to comply with. A script would delete everything we could from the DB immediately. Since we had 30 day backups, their data would only be deleted from the backups on day 31. I think this was acceptable to the GDPR consultant.
Going in to the backups to delete their data there in insane.
If I was legally obliged to delete data then I'd make sure I deleted, regardless of the purpose or location of the storage. If you can't handle a delete request you shouldn't collect the data in the first place.
If you don't want your data online then don't put it there.
But this obviously means building an extensive system to ensure the encoded identifier is the only thing used across your system (or a giant key management system).
In the past I’ve been a part of systems at exabyte scale that had to implement this. Hard but not impossible. I can see how orgs try to ‘legalese’ their way out of doing this though because the only forcing function is judicial.
Nevertheless, you're generally correct but you don't realize why: A core feature of ChatGPT is that it keeps your conversation history right there so you can click on it, review it, and continue conversations across all of your devices. The court order is to preserve what is already present in the system even if the user asks to delete it.
For those who are confused: A core feature of ChatGPT and other LLM accounts is that your past conversations are available to return to, until you specifically delete them. The problem now is that if a user asks for the conversation to be deleted, OpenAI has to retain the conversation for the court order even though it appears deleted.
Is it a requirement that ChatGPT users own multiple computers
Is it a requirement that ChatGPT users use ChatGPT on multiple computers
Is it true that a goal of online advertising services providers is to learn about all of an ad targets' computers and link them to a single identity
Is every software "feature" necessary
Are there "features" in some software that benefit software developers more than software users, e.g., through data colllection, surveilllance and advertising services
Should all software "features" chosen by developers be "opt-out", with default settings chosen by developers not users, or should some be "opt-in"
What if a "feature" chosen by a developer that no user ever requested cannot be implemented as "opt-in". Should users that do not wish to subject themselves to the "feature" use the software
Is ChatGPT chat history a "feature"
Your previous ChatGPT conversations show up right in the ChatGPT interface.
They have to store the private conversations to enable users to bring them up in the interface.
This isn't a secretive, hidden data collection. It's a clear and obvious feature right in the product. They're fighting for the ability to not retain secret records of past conversations that have been deleted.
The problem with the court order is that it requires them to keep the conversations even after a user presses the 'Delete' button on them.
OpenAI made a choice in how the feature was and is implemented.
The whole point of ChatGPT conversations is so they can be read by the model on the server.
Conversations are kept around because they can be picked up and continued at any point (I use this feature frequently).
Additionally you can use conversations in their scheduled notification feature, where the conversation is replayed and updates are sent to you, all done on the server.
> OpenAI made a choice in how the feature was and is implemented.
Indeed they did, and it was a sensible choice given how the conversations are used.
Models should run in ephemeral containers where data is only processed in RAM. For active conversation a unique and temporary key-pair is generated. Saved chats are encrypted client side and stored encrypted server side. To resume a conversation[0], decrypt client side, establish connection to container, generate new temporary key-pair, and so on. There's more details and nuances but this is very doable.
How Mullvad handles your data, for some inspiration: https://mullvad.net/en/help/no-logging-data-policy
> Conversations are kept around because they can be picked up and continued at any point (I use this feature frequently).
> Additionally you can use conversations in their scheduled notification feature, where the conversation is replayed and updates are sent to you, all done on the server.
Of course, I also don't see OAI pushing the state of Homomorphic Encryption forward either... But there's definitely a lot of research and more than acceptable solutions that allow data to be processed server side while being encrypted for as long as possible and making access to that data incredibly difficult.
Again, dive deep into how Mullvad does it. It is not possible for them to make all their data encrypted, but they make it as close to impossible to get, including by themselves. There doesn't need to be a perfect solution, but there's no real reason these companies couldn't restrict their own access to that data. There's only 2 reasons they are not doing so. Either 1) they just don't care enough about your privacy or 2) they want it for themselves. Considering how OpenAI pushes the "Scale is All You Need" narrative, and "scale" includes "data", I'm far more inclined to believe the reason is option 2.
[0] Remember, this isn't so much a conversation in the conventional sense. The LLMs don't "remember". You send them the entire chat history in each request. In this sense they are Markovian. It's not like they're tuning a model just to you. And even if they were, well we can store weights encrypted too. Doesn't matter if a whole model, LoRA, embeddings, or whatever. That can be encrypted at rest via keys OAI does not have access to.
Mullvad and E2EE Messengers do not need to process the contents of the message on their server. All they do is, passing it to another computer. It could be scrambled binary for all they care. But any AI company _has_ to read the content of the message by definition of their service.
Maybe, but letting aside that they are two different kind of products, how can you trust them to really do so? And in any way, in the case of ChatGPT where should I store my client side private key, as I use those bots only in my web browser? Maybe in my password manager and I copy paste it every time I start a new conversation.
My take is that if they went this way we would not be talking about them now, we would be talking about one of their competitors that didn't put hurdles between their product and their customers.
In other words, survivor bias.
I built E2E encrypted LLMs using secure enclaves, so I know a bit about this space.
The tech works, for small LLMs - the sort of thing you can run on your mobile already. It isn't yet (?) there for LLMs the size of ChatGPT.
- [1] https://arstechnica.com/tech-policy/2025/08/openai-offers-20...
There is nothing in the article that mentions OpenAI being forced to create new logs they don't already have.
This is true in services like Datadog, New Relic, and logging services like Splunk. But even privacy-focused services like Mullvad keep logs for 24 hours to monitor for abuse. So this concept that retaining logs is significantly weaker than not ordering the collection is really a bit of misdirection. I'm not sure whether it's intentional, but it's definitely misleading.
It’s not misdirection or misleading; it lies in an understanding of the law. There’s plenty of case law out there on the subject if you’re interested.
So in this case not keeping logs as ordered by the court would be contempt of court.
There is a separate higher-tier private API customers can pay for that never had logging enabled, and the court did not force the company to add it.
Its needed for the conversation history feature, a core feature of the ChatGPT product
Its like saying "What is the purpose of Google Photos storing millions of private images"
Have you used ChatGPT? Your conversation history is on the left rail
I can search the logs I keep without decompressing
Every user is different and each is free to use whatever software they want
No
Large number of upvotes on the quoted comment however. Maybe some of those voters are ChatGPT users
I do searching from the command line in text mode. The script I use keeps a "log" (a customised SERP) of all query strings and search result URLs. I also have these URLs stored in the logs from the forward proxy. These are compressed using RePair. I can search the compressed logs faster this way than with something like
ztsd -dc log.zst|grep pattern
rg -z pattern log.zstGiven that, I'd suggest not offering "alternatives" to the features described in TFA for a service you've never used. There are people here talking about oranges, a lot of them with domain expertise, and you're not just talking about apples, you're talking about bird migrations.
Okay well it's a chat app where you chat directly with an LLM. The way LLMs work is you feed the entire chat history into it, and it generates the next message. Therefore, there's no way you can chat with it without storing the history. It's impossible
Sure, and also downvotes - that measures factionalism, not correctness.
But tech wise, you're confused. Functionally speaking chatgpt is a shared document editor - the server needs to store chat histories for the same reason Google Docs stores the content of documents. Users can submit text to chatgpt.com from one browser, and later edit that text from the app or a different browser. Ergo the text is stored on the server, simple as that.
3 versus 190+, so far
Many commenters cannot distinguish rhetorical questions from questions that seek an answer
By attempting to answer a rhetorical question one may only strengthen the point being made by the question, for example, poor decision-making, and may reveal an absence self-awareness
To do this, I first insert a blank line at the top of each log file before adding to the tarball
IME, RePair is faster than compressing with zstd and the size reduction is almost the same
The only "catch" is that RePair requires more memory during compression
Unfortunately, different searches for this RePair you mentioned have only revealed links to resources for repairing broken air compressors, damaged compressed files, spinal injuries, etc.
Instead of writing open letters they could of course do something about it. Even Google stopped storing your location timeline on their servers and now have it per-device only.
In fact, as far as I understand it, they could not be made public by the plaintiffs even if they wanted to do so, or even if one of their employees decided to leak them.
That's because the plaintiffs themselves never actually see the documents. They will only be seen by the plaintiff's lawyers and any experts hired by those lawyers to analyze them.
https://ia801205.us.archive.org/1/items/gov.uscourts.nysd.61...
OpenAI October 30, 2025 Letter Opposing Motion to Compel
https://ia601205.us.archive.org/1/items/gov.uscourts.nysd.61...
November 7, 2025 Order on Motion to Compel
https://ia601205.us.archive.org/1/items/gov.uscourts.nysd.61...
"OpenAI has failed to explain how its consumers privacy rights are not adequately protected by: (1) the existing protective order in this multidistrict litigation or (2) OpenAIs exhaustive de-identification of all of the 20 million Consumer ChatGPT Logs.1
1. As News Plaintiffs point out, OpenAI has spent the last two and a half months processing and deidentifying this 20 million record sample. (ECF 719 at 1 n.1)."
https://ide.mit.edu/wp-content/uploads/2018/01/w23815.pdf
Companies operating popular www search engines might claim that the need for longer retention is "to provide better service" or some similar reason that focuses on users' interests rather than the company's interests^2
2. Generally, advertising services
This paper attempts to expose such claims as bogus
1. According to some reports OpenAI is sending some queries to Google
For users affected by OpenAI's conduct, an "answer" makes no difference. Anyone can construct any "answer" they want and we can see that in this thread. For users affected by OpenAI's conduct, it does not matter
In the above paper on search engines, the claim was that longer retention of sensitive data leads to better search. This was the "answer" presented in response to the question of "why"
But the "answer" is only misdirection. The companies have no reputation for being honest and their operations are non-transparent. Accordingly, user focus will be on the consequences for users of the company's practices, not "why"
Some readers are probably too young to have read through the AOL search data
https://en.wikipedia.org/wiki/AOL_search_log_release
Did anyone care "why" AOL released the data
IMHO, it is unfortunate that papers like the one above need to published
The question of "why" is rhetorical. It is meant to the draw attention to the consequences for users, not to seek an "answer"
HN replies might try to answer this as well but the answer is already known to the world
The conversations will be made available to the plaintiffs' (including New York Times') attorneys and the plaintiffs' attorneys' experts
If OpenAI did not store such conversations as a matter of practice before being sued, then there would be no private conversations to make available to the plaintiffs' attorneys and their experts
275 upvotes
AFAICT, most HN readers did _not_ misintepret the question
HN replies != HN, it is a small subset of the readership
To train the AI further. Obviously. Simple as.
Why do chat histories need to be accessible by OpenAI, its service partners and anyone with the authority to request them from OpenAI
If users want this design, as suggested by HN commenters, if users want their chat histories to be accessible to OpenAI, its service providers and anyone with authority to request them from OpenAI, then wouldn't it also be true that these users are not much concerned with "privacy"
If so, then why would OpenAI proclaim they are "fighting the New York Times' invasion of user privacy", knowing that NYT is prohibited from making the logs public and users generally do not care much about "privacy" anyway
The restrictions on plaintiff NYT's use of the logs are greater than the restrictions, if any,^1 on OpenAI's use of them
1. If any such restrictions existed, for example if OpenAI stated "We don't do X" in a "privacy policy" and people interpreted this as a legally enforceable restriction,^2 how would a user verify that the statement was true, i.e., that OpenAI has not violated the "restriction". Silicon Valley companies like OpenAI are highly secretive
2. As opposed to a statement by OpenAi of what OpenAI allegedly does not do. Compare with a potentially legally-enforceable promise such as "OpenAI will not do X". Also consider that OpenAI may do Y, Z, etc. and make no mention of it to anyone. As it happens Silicon Valley companies generally have a reputation for dishonesty
OpenAI could store the chat conversation in an encrypted format that only you, the user, can decrypt, with the client-side determining the amount of previous messages to include for additional context, but there's plenty of user overhead involved in an undertaking like that (likely a separate decryption password would be needed to ensure full user-exclusive access, etc).
I'd appreciate and use a feature like that, but I doubt most "average" users would care.
Ever since I put the effort into Syncthing across my all devices (paired with restic on one of them for backup), I can't help but see how cross-device functionality and cloud this are the Sysco hash potatoes that balloons Big Corp services' profit margins.
Not saying it's easy to set up. But when you get there it's so liberating and you wish all software was bring-your-own-network.
Am I missing something? Is this seriously a heated HN debate over "why does this website need to store the text it sends to people who view the website?"?
And the clients don't need to be running at the same time if you have a third device that's always on and receiving the changes from either (like a backup system). Eventually everything arrives. It's not as robust as what Google or iCloud gives you, but it's good enough for me.
I absolutely want OpenAI to keep all of my chats and I absolutely don't want them to share them ( voluntarily or by force) with any private agent.
I have exactly the same expectation of any document or communication platform. It's been long established as accepted compomise between security and convenience.
People access ChatGPT through different interfaces: Web, desktop app, their phones, tablets.
Therefore the conversations are stored on the servers. It's really not some hidden plot against users to steal their data. It's just how most users expect their apps to work.
That's a tricky system to get right and maintain
(Please do not interpret this as a defense of OpenAI! I just think that we shouldn't trivialize the task of encrypting user data so that it's not visible to the provider).
If you read the article, you'd see this:
> Our long-term roadmap includes advanced security features designed to keep your data private, including client-side encryption for your messages
Open Ai didn't want to.
For example, I save all the POST request bodies I send over the internet in the local forward proxy's log. I add logs to tarballs and compress with an algorithm that allows for searching the logs in the tarballs without decompressing them
It does not matter what "reason" or "excuse" or "explanation" anyone presents, technical or otherwise, for why OpenAi does what it does
The issue is what are the consequences
If you have ever played with the api, its clear as day that the protocol itself is stateless.
As might any plaintiff. NYT might be the first of many others and the lawsuits may not be limited to copyright claims
Why has OpenAI collected and stored 20 million conversations (including "deleted chats")
What is the purpose of OpenAI storing millions of private conversations
By contrast the purpose of NYT's request is both clear and limited
The documents requested are not being made public by the plaintiffs. The documents will presumably be redacted to protect any confidential information before being produced to the plaintiffs, the documents can only be used by the plaintiffs for the purpose of the litigation against OpenAI and, unlike OpenAI who has collected and stored these conversations for as long as OpenAI desires, the plaintiffs are prohibited from retaining copies of the documents after the litigation is concluded
The privacy issue here has been created by OpenAI for their own commercial benefit
It is not even clear what this benefit, if any, will be as OpenAI continues to search for a "business model"
Wanton data collection