Osegi
Security·1 min read

ChatGPT Flaw Let a Shared Chat Link Expose User Histories

Researchers at Parameter found a ChatGPT vulnerability in which a shared-conversation link could run malicious code and reach a user's chat history. OpenAI has fixed it.

A laptop screen showing a chat interface with a red warning padlock symbolizing a security flaw in ChatGPT.
Image: Antyweb

A vulnerability in ChatGPT allowed an ordinary link to a shared conversation to be turned into a tool for stealing private user data. Researchers at the firm Parameter discovered the flaw, which OpenAI has since fixed.

How the attack worked

According to the researchers, the bug made it possible to run malicious code inside the active session of a logged-in user. Once the code was running, an attacker could gain access to that person's conversation history.

The entry point was deliberately mundane: a link to a shared chat, the kind of URL people routinely send to colleagues or post online. That is what made the weakness worrying, since opening such a link is something most users would not think twice about.

Why it matters

ChatGPT histories often contain sensitive material, from personal questions and work documents to business plans and code. Access to a whole history could therefore expose far more than a single exchange.

OpenAI has patched the issue, so the vulnerability no longer works in its original form. The researchers stress, however, that the potential consequences were serious.

The case is a reminder that features designed for convenience, such as sharing conversations, widen the attack surface of AI assistants. Users who keep sensitive information in their chats may want to review what they store there and be cautious with links from unknown sources.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:chatgptopenaivulnerabilityparametershared chatsprivacy

More from this day

See all →

More in: Security

See all →