CHATGPT OPERATOR PROMPT INJECTION EXPLOIT LEAKING PRIVATE DATA

Summary from the TLDR InfoSec newsletter

ChatGPT Operator is an advanced AI agent equipped with web browsing and reasoning capabilities that can interact with websites on behalf of users. It is vulnerable to Prompt Injection attacks via instructions embedded into text or web content that can cause the model to browse sensitive pages and leak the information to attacker-controlled websites. OpenAI has implemented defenses such as prompting the user for confirmation, but researchers have shown that these mitigations can be bypassed.