OpenAI has introduced a new Lockdown Mode designed to enhance protection against prompt injection attacks. These attacks involve embedding malicious instructions for chatbots within web pages and other content sources. The Lockdown Mode will disable live web browsing, limiting access to cached content only. Additionally, it will block the retrieval and display of images from the web, although users can still generate images. The functionality for deep research and agent mode will also be restricted. Despite these measures, OpenAI warns that ChatGPT could still be susceptible to prompt injections from cached web content or uploaded files, potentially impacting the accuracy or behavior of its responses. The main goal of Lockdown Mode is to minimize the risk of inadvertently sharing sensitive data. OpenAI emphasizes that this mode is specifically intended for individuals and organizations dealing with sensitive data and seeking increased protection against data exfiltration risks linked to prompt injections. Lockdown Mode is being rolled out to self-serve ChatGPT Business accounts and eligible personal accounts.
How will OpenAI’s Lockdown Mode protect against prompt injection attacks?
