**In Short.**
Posted: Je ne suis pas tout à fait sûr de ce que vous voulez dire. OpenAI has unveiled a new feature designed to offer extra defense against prompt injection attacks, in which harmful instructions for chatbots are concealed within webpages and other data sources. Among its effects, Lockdown Mode will turn off live web browsing (limiting access to cached content only), block the retrieval and display of images from the internet (while still allowing image generation), and disable deep research and agent mode. The company notes that even when Lockdown Mode is active, ChatGPT may remain susceptible to prompt injections—for instance, those embedded in cached web pages or uploaded files—which could still influence the behavior or accuracy of its responses. The objective is to minimize the chances of sensitive information being shared during the process. “Lockdown Mode is not intended for everyone,” OpenAI says. It is intended for individuals and organizations managing sensitive data who seek enhanced safeguards against data exfiltration threats stemming from prompt injection. The company is now deploying Lockdown Mode to self-serve ChatGPT Business accounts and eligible personal accounts.
