The P1 flaw reportedly let malicious shared links execute attacker code in signed-in ChatGPT sessions.
Malicious shared links could reportedly run attacker code in a signed-in ChatGPT session and access private conversations. OpenAI reportedly fixed the P1 flaw.
188
5 Sources, first seen ago
@thecircuitry_Researchers bypassed ChatGPT's sandbox and content policy to hijack accounts with a shared link. The flaw let attackers read private chats in the victim's session before OpenAI fixed it on October 26.
@signalfoldintelParameter showed that a crafted ChatGPT share link could read private chats through the viewer's signed-in session. The researchers say OpenAI fixed the flaw on 26 Oct. Sandboxed previews still need authenticated cross-frame messaging. https://www.parameter.ai/labs/we-found-a-gap-at-openai@thecircuitry_Researchers bypassed ChatGPT's sandbox and content policy to hijack accounts with a shared link. The flaw let attackers read private chats in the victim's session before OpenAI fixed it on October 26.@signalfoldintelParameter showed that a crafted ChatGPT share link could read private chats through the viewer's signed-in session. The researchers say OpenAI fixed the flaw on 26 Oct. Sandboxed previews still need authenticated cross-frame messaging. https://www.parameter.ai/labs/we-found-a-gap-at-openai@thecircuitry_Researchers bypassed ChatGPT's sandbox and content policy to hijack accounts with a shared link. The flaw let attackers read private chats in the victim's session before OpenAI fixed it on October 26.@signalfoldintelParameter showed that a crafted ChatGPT share link could read private chats through the viewer's signed-in session. The researchers say OpenAI fixed the flaw on 26 Oct. Sandboxed previews still need authenticated cross-frame messaging. https://www.parameter.ai/labs/we-found-a-gap-at-openai@thecircuitry_Researchers bypassed ChatGPT's sandbox and content policy to hijack accounts with a shared link. The flaw let attackers read private chats in the victim's session before OpenAI fixed it on October 26.@signalfoldintelParameter showed that a crafted ChatGPT share link could read private chats through the viewer's signed-in session. The researchers say OpenAI fixed the flaw on 26 Oct. Sandboxed previews still need authenticated cross-frame messaging. https://www.parameter.ai/labs/we-found-a-gap-at-openai—
Not ranked yet
—
Not ranked yet
