Cybersecurity researchers have disclosed details of a vulnerability in OpenAI ChatGPT. This flaw leverages the artificial intelligence (AI) assistant's implicit trust in Markdown links and images. It can trigger prompt injections and open the door to phishing attacks. The technique has been codenamed ChatGPhish by Permiso Security. This highlights the specific nature of the exploit. The core issue is that the chatgpt.
Cybersecurity researchers have disclosed details of a vulnerability in OpenAI ChatGPT. This flaw leverages the artificial intelligence (AI) assistant's implicit trust in Markdown links and images. It can trigger prompt injections and open the door to phishing attacks.
The technique has been codenamed ChatGPhish by Permiso Security. This highlights the specific nature of the exploit.
The core issue is that the chatgpt.com response renderer implicitly trusts Markdown links and Markdown formatting.