source: techcrunch ai: openai unveils lockdown mode to protect sensitive data from prompt injection attacks
level: technical
openai announced a new feature called lockdown mode to reduce risks from prompt injection attacks. these attacks hide malicious instructions in webpages or other content that can trick chatbots into leaking sensitive data. lockdown mode disables live web browsing, so only cached content is available. it also stops the retrieval and display of images from the web, though image generation still works. deep research and agent mode are turned off too.
the company warns that even with lockdown mode on, chatgpt could still be vulnerable. prompt injections might appear in cached web content or uploaded files and could affect responses. the main goal is to lower the chance that sensitive data gets shared during these attacks. openai says this feature is not for everyone. it is meant for people and organizations handling sensitive data who want stricter protection from data exfiltration risks tied to prompt injection.
lockdown mode is rolling out now to self-serve chatgpt business accounts and eligible personal accounts. this move addresses growing concerns about security in ai systems that interact with external content. by limiting dynamic web access, openai aims to create a safer environment for users dealing with confidential information. the feature reflects ongoing efforts to balance functionality with security in large language models.
why it matters: it helps ai practitioners and businesses reduce the risk of sensitive data leaks when using chatbots that process untrusted content.
source: techcrunch ai: openai unveils lockdown mode to protect sensitive data from prompt injection attacks