OpenAI's Lockdown Mode cuts agents to stop prompt injection
openai launched lockdown mode for business accounts to prevent prompt injection data exfiltration by disabling web browsing and agent capabilities.
openai released lockdown mode, a new security feature for chatgpt business and eligible personal accounts designed to thwart prompt injection attacks where malicious instructions are hidden in web content or files. to reduce data exfiltration risks, the mode disables live web browsing, web image retrieval, deep research, and agent mode. openai acknowledges that injections in cached content or uploads remain a risk but aims to significantly lower the probability of sensitive data leaks.