
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Researchers at Check Point discovered a vulnerability in ChatGPT's Lockdown Mode that allowed attackers to bypass security controls, enabling silent theft of chat data, phishing, and malware delivery. The flaw, rooted in HTML preview handling, was promptly fixed by OpenAI, but highlighted broader AI security challenges.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves an AI system (ChatGPT) and its security controls being bypassed, enabling prompt injection attacks that could lead to data theft, phishing, and malware delivery. These harms fall under injury or harm to persons and communities. The researchers demonstrated the exploit, showing the AI system's malfunction or misuse directly or indirectly leading to potential harm. OpenAI's patching is a response but does not negate the incident itself. Therefore, this is an AI Incident due to the realized vulnerability and demonstrated potential for harm.[AI generated]