Why ‘Temporary’ Chats Aren’t So Temporary: OpenAI, Court Orders, and Your Forever-Archived Prompts

You probably thought “temporary” meant temporary. Cute. In a recent Reddit thread, a user pointed out something many of us blissfully ignored: because of an ongoing court order tied to the New York Times lawsuit, OpenAI has paused deletions and is preserving chats that could be relevant to the case. That includes normal chats, chats you thought you deleted, the 30-day “temporary” conversations, and voice dictation. In short: if you told ChatGPT something in confidence, it may be enjoying an extended vacation in a legal vault until the litigation is resolved.

The technical and legal fine print is buried in OpenAI’s Terms of Service: the company can “preserve or disclose your information if we believe it is reasonably necessary to comply with a law, regulation, legal process, or governmental request.” That clause is not new. What changed is a court order that treats deleting potentially relevant chats as the risk of destroying evidence, so OpenAI has been instructed to retain data that might matter to the case until the judge says otherwise.

Redditers were predictably split between people who already knew and sighed, and others who were surprised that the “delete” button is more cosmetic than magical. A few high-value comments clarified the situation: one pointed to the NYT lawsuit and the judge’s preservation demand; another reminded readers that “delete” often just flips a database flag rather than truly erasing data; and a privacy advisor offered a short GDPR primer noting that deletion rights can be lawfully paused for compliance with legal obligations (see GDPR Article 17 exceptions).

Why this matters

A court-ordered preservation has both practical and philosophical consequences. Practically, data that would have auto-vanished after 30 days can now be retained, which raises storage, access, and governance questions. Philosophically, it forces a reckoning over what “delete” and “privacy” actually mean when private speech intersects with litigation and corporate practices.

From a user perspective, the immediate implication is simple: assume your prompts and voice notes can be retained and potentially reviewed if a court deems them relevant. From a regulatory perspective, the case could set precedent on how AI companies balance user privacy, evidence preservation, and the costs of long-term retention. From a product perspective, it is a headache—preserving billions of chats is expensive and not how a fast-scaling startup planned to spend its storage budget.

What the community said (high-value comments)

• Key-Balance-9969: Points to the NYT lawsuit as the root cause. The judge’s order treats chats as potential evidence, which explains the blanket preservation requirement.
• neuro__atypical: A reminder that “delete” often equals a hidden database flag, not a digital vanishing act. Useful reality check for anyone who treats internet delete buttons as magic.
• FiveNine235: Provides GDPR context. Yes, deletion rights can be temporarily suspended if there is a lawful obligation to retain data. The advisor also notes OpenAI has set up limited-access storage to comply with the order while attempting to minimize broader access.

What you can do

1. Don’t feed sensitive secrets into cloud LLMs. Treat prompts like semi-public notes. If you would not text it to a stranger, do not paste it into a prompt.
2. Consider local models for private work. Tools like Ollama, LlamaIndex, and other on-prem or local LLM setups can reduce exposure when privacy matters.
3. If you are in the EU, you can still exercise GDPR rights, but know they may be lawfully suspended while preservation orders are active. Keep records of requests and monitor developments.

The takeaway, if there is one, is that “temporary” now comes with caveats. OpenAI says it is complying with a court order, which is not the same as proactively hoarding user data, but the practical effect feels similar for users worried about privacy. For now, treat cloud LLMs as persistent recording devices and design your behavior around that reality. And yes, maybe reread the Terms of Service next time before confessing your weirdest internet searches to an algorithm.