Sam Altman calls for ‘AI privilege’ as OpenAI clarifies court order to retain temporary and deleted ChatGPT sessions

Ordinary ChatGPT users (including the author of this article) may notice that Chatbot from Openai allows users to enter a “temporary chat” designed to survey all mutual information between the user and the basic artificial intelligence model once the chat session is closed.

In addition, Openai also allows users to delete the previous ChatgPT sessions from the left sidebar on the web and desktop/mobile phone applications by left or click control, or press/press the long from the specified.

However, Openai this week was found to face criticism from some Chatgpt users after discovering that the company has no In fact, these chat records were deleted as shown previously.

“You tell me that my deleted chats have not been deleted in reality and [are] It is rescued for investigation by the judge? X User @ns123abc has been published.

Another user added, Kepano, “You can” delete “chat chat, yet all the chat should be kept due to legal obligations?”

Simon Willison also wrote AI and software engineer on his personal blog: “Pay customers [OpenAI’s] Application programming facades may make a decision to switch to other service providers who can provide retaining policies that have not been sabotaged according to this court order! “

Instead, Openai confirmed that deleted and temporary user chat records have been preserved since mid -May 2025 in response to the Federal Court order, although it did not reveal this to users until June 5.

It takes, the built -in below and its release on May 13, 2025, the American judge judge ONA T

Court stems from The New York Times (NYT) against Openai and MicrosoftThe issue of old copyright throughout the year and a half is still in debate. the NewtonS. Lawyers claim that Openai’s language models are re -finished with the content of copyright. Prosecutors argue that the records, including those that users may delete, may contain an output of the case.

Although Openai complied with the demand immediately, it did not notify users who are publicly affected for more than three weeks, as he issued a blog post and common questions describing the legal mandate and determining the affected lines.

However, Openai puts the blame directly on Newton The judge ordered, saying that he believed that the request for memorization was “baseless.”

Openai shows what is going on by the court order to maintain the records of the ChatgPT user – including the affected chats

In a blog post published yesterday, Openai Coo Brad Lightcap motivated the company’s position and stated that he was defending the privacy of the user and safety against an excessive judicial order, writing:

“The New York Times and other prosecutors have submitted a comprehensive and unnecessary request in an unfounded lawsuit from our department against us: keeping the consumer customer and API data indefinitely. This is mainly inconsistent with the privacy obligations that we made on our users.”

The post explained that Chatgpt free users, in addition to the Pro, Pro and Team, along with API customers without the ZDR retaining agreement, are affected by the memorization matter, which means even if users in these plans delete their chat or use temporary chat mode, their conversations will be stored in the distinguished future.

However, the subscribers of ATCGPT Enterprise and EDU, as well as API customers using the end of ZDR points, are no It is affected by the order and their chats are deleted as directed.

The detained data is kept under a legal suspension, which means that it is stored in a safe and detailed system and can only be accessed to a small number of legal and safety personnel.

“These data are not shared automatically New York Times Or anyone else, “Lightcap confirmed in the Openai Blog post.

Sam Germany floats a new concept of “artificial intelligence concession” to allow secret conversations between models and users, similar to speaking to a human doctor or a lawyer

The CEO of Openai and co -founder Sam Altman publicly dealt with the issue in a post from his account on the social network X last night, and writing:

“Recently, NYT asked the court to force us not to delete any user conversations. We believe this was an inappropriate request to put a bad precedent. We appeal to the decision. We will fight any request that fights the privacy of our users; this is a basic principle.”

He also suggested that there is a broader legal and moral framework for the privacy of artificial intelligence:

“We have recently thought about the need for something like” artificial intelligence privilege “; this really accelerates the need to convey.”

“Talking to artificial intelligence should be like talking to a lawyer or a doctor.”

“I hope this community will discover soon.“

The idea of ​​artificial intelligence concession-a potential legal standard-is frequent in the customer of the customer and the secret of the doctor and the patient.

Whether such a framework will get attracted in courtrooms or political circles, it still must be seen, but Altman’s statements indicate that Openai has increasingly defending such a shift.

What comes after that for Openai and temporary/deleted conversations?

Openai submitted an official protest against the court’s order, and asked to evacuate it.

In the court files, the company argues that the request lacks a realistic basis and that maintaining billions of additional data points is neither necessary nor proportional.

Judge Wang indicated, in a hearing on May 27, that the matter is temporary. The parties have ordered the development of the sampling plan to test whether the deleted user data is financially different from the detained records. Openai was ordered to make this proposal by today (June 6), but I haven’t seen the deposit yet.

What it deals with institutions and decision makers responsible for using ChatGPT in corporate environments

While the demand exempt Chatgpt Enterprise and API customers using the end of ZDR, the most important legal effects and reputation are important for professionals responsible for spreading and expanding artificial intelligence solutions within organizations.

Those who oversee the full life cycle of LLMS models-from swallowing data to control and integration-will need to re-evaluate assumptions on data governance. If the components facing a LLM user are subject to legal memorization orders, they raise urgent questions about the place where the data leaves after it leaves a safe end point, how to isolate, register or hide high -risk reactions.

Any platform to touch the Openai application programming facades must verify the validity of the end points (ZDR VS Non-ZDR) and make sure that data processing policies are reflected in user agreements, audit records and internal documents.

Even if the end of ZDR points are used, the data cycle life policies may require review to ensure that the escort systems (analyzes, registration, backup) do not inadvertently retain the temporary interactions that were short -lived.

Security personnel responsible for risk management must now expand the modeling of threats to include the legal discovery as a potential head. The teams must verify whether the practices of retaining the back interface of Openai corresponds to internal controls and risk assessments of external parties, and whether users depend on features such as “temporary chat” that no longer work as expected in the field of legal memorization.

A new flash point for the privacy of the user and safety

This moment is not just legal skirmishes. It is a flash point in advanced conversation on the privacy of artificial intelligence and data rights. By framing the issue as the issue of “artificial intelligence concession”, Openai is effectively suggested a new social contract for how to deal with smart systems of secret inputs.

Whether the courts or legislators accept that the framework is still not certain. But at the present time, Openai is arrested in a budget law – between legal compliance, institutions assurances and user confidence – and facing questions with a louder voice about who controls your data when speaking to the device.