OpenAI rolls out ChatGPT Lockdown Mode for prompt-injection risks (opens in new tab)

OpenAI has begun rolling out Lockdown Mode for ChatGPT, a security setting designed to reduce the risk that prompt-injection attacks cause sensitive data to be sent outside a user's chat session. Prompt-injection attacks hide malicious instructions in sources such as web pages, comments, reviews, cached pages or uploaded files, which a model may treat as instructions instead of ordinary content. The mode limits ChatGPT's internet-connected functions: reports said it disables or restricts live...