AI Issue: ChatGPT Privacy Enhancement Preview
As privacy concerns grow, OpenAI is reportedly considering encryption for ChatGPT, starting with temporary chats.
Key Highlights
- OpenAI is considering encryption for ChatGPT, starting with temporary chats.
- Temporary chats don't appear in history or training models, but they're still subject to a federal court order requiring preservation.
- Full end-to-end encryption is a complex problem for AI systems since providers like OpenAI serve as endpoints and need access to user data for features like long-term memory.
- Altman said users sharing sensitive medical or legal issues are demanding protections similar to doctor-patient or attorney-client privilege.
- While law enforcement requests are currently rare (double digits annually), a single major case could accelerate change, Altman predicted.
Encryption Plans for ChatGPT
OpenAI CEO Sam Altman said the company is thinking "very seriously" about bringing encryption to ChatGPT, though no launch timeline has been set.
Altman suggested temporary chats would be the most likely starting point.
These chats currently don't appear in user history and aren't used to train models, but OpenAI can retain copies for up to 30 days for safety purposes.
Still, temporary or deleted chats are subject to a federal court order issued last May requiring OpenAI to preserve their contents.
This legal backdrop complicates the push for encryption, since data can still be compelled through the courts.
Altman said watching how ChatGPT is actually used has highlighted the growing importance of people sharing sensitive data:
"People are pouring their hearts out to ChatGPT about their most sensitive medical issues and everything else. That has led me to believe that AI privacy is extremely important."
Unlike conversations with doctors or lawyers, which are protected by legal privilege, AI chats currently lack comparable safeguards.
The Difficulty of AI Encryption
Encryption is straightforward in messaging apps where only the endpoints hold the keys.
But with AI chatbots, the provider (in this case OpenAI) is a party to the conversation.
This means simply encrypting data in transit doesn't prevent OpenAI from accessing chat content or sharing it with law enforcement.
True end-to-end encryption is therefore difficult to implement.
Some companies like Apple have introduced partial solutions.
For example, Private Cloud Compute for Apple Intelligence allows queries to run on Apple servers without broadly exposing data to the company.
The challenge for OpenAI is much greater.
Features like ChatGPT's long-term memory require the company to maintain access to user data.
Fully encrypting everything would require disabling or redesigning those features, creating a tradeoff between privacy and functionality.
Government Access and Growing Pressure
Altman acknowledged that OpenAI hasn't received many data requests from law enforcement yet:
"It's still a very low number, double digits per year, but it's growing. And I think one really big case is going to happen, and people are going to say, 'you know, we really need a different approach here.'"
He added that the risk has increased as users now treat ChatGPT like a trusted friend for medical or legal matters.
These conversations are similar to professional consultations with doctors or lawyers, but using AI lacks those protections.
"If they can get a better version of [medical and legal chats] from AI, they should be able to get the same protections for the same reasons," Altman says.
Altman noted that this issue was originally not on his radar but became a priority after realizing the scale of sensitive data people share.
Q&A
Q: What type of encryption is OpenAI considering for ChatGPT?
A: OpenAI is exploring encryption starting with temporary chats, which are not saved to history or used for model training.
Q: Why are temporary chats the most likely first step?
A: They already operate with limited preservation of up to 30 days for safety, but are subject to federal court orders requiring preservation.
Q: Why is full end-to-end encryption difficult for ChatGPT?
A: Because OpenAI itself is an endpoint of the conversation, and features like long-term memory require access to user data.
Q: How often does OpenAI receive data requests from law enforcement?
A: Altman says it's still very low at double digits per year, but growing.
Q: Why does Altman think encryption is urgent now?
A: He said users are sharing sensitive medical and legal issues with ChatGPT, which has led him to believe AI privileges and stronger protections are needed.
Implications
The encryption debate around ChatGPT highlights a growing tension: users are increasingly relying on AI for sensitive and important conversations, yet these chats currently lack the legal protections given to human professionals.
Altman's comments reveal a shift within OpenAI where privacy is moving from a secondary concern to a front-and-center issue.
Altman and OpenAI have begun pushing for new safeguards, arguing that AI consultations should receive the same privileges as doctor-patient or attorney-client conversations, including protection from government access.
He predicted that lawmakers, who have generally been receptive to privacy measures, will eventually adopt such protections.
"I don't know how long it'll take," Altman said. "I think society has to evolve."
While full end-to-end encryption remains technically difficult, starting with temporary chats could be an important first step.
The bigger question is whether AI companies, courts, and lawmakers can agree on a model where privacy, safety, and functionality coexist.
IMPAKERS Blog | GPT-5 Release Issue Read More
Source: Alicia Shapiro, AiNews, "Encryption Could Be Coming to ChatGPT, Says Sam Altman", https://www.ainews.com/p/encryption-could-be-coming-to-chatgpt-says-sam-altman, (2025.08.18)