OpenAI unveils Lockdown Mode to protect sensitive data from prompt injection attacks

OpenAI has launched an innovative feature aimed at enhancing security against prompt injection attacks—a tactic where harmful instructions are concealed within websites and various content sources. This new Lockdown Mode offers crucial safeguards by disabling live web browsing, restricting access to only cached content, and preventing the retrieval and display of online images, although users can still create images independently. Despite these measures, OpenAI cautions that even with Lockdown Mode activated, ChatGPT might remain susceptible to prompt injections. Such vulnerabilities could stem from cached web content or uploaded files, potentially influencing the accuracy and behavior of responses. However, the primary objective of this feature is to minimize the risk of inadvertently sharing sensitive information. OpenAI emphasizes that Lockdown Mode is specifically tailored for individuals and organizations that manage sensitive data and require enhanced protections against data exfiltration risks linked to prompt injection. The rollout of Lockdown Mode is currently underway for self-serve ChatGPT Business accounts, alongside select personal accounts.