OpenAI has introduced a new security feature called Lockdown Mode for ChatGPT to help protect users and organisations from prompt injection attacks and reduce the risk of sensitive data theft. The optional setting is now available across all ChatGPT account types, including personal, business, and enterprise workspaces. When activated, Lockdown Mode restricts several internet-dependent features in the AI chatbot to limit network requests that could be exploited to steal data.
Prompt injection attacks occur when malicious instructions are embedded in web pages, files, or documents fed to the AI, potentially tricking it into revealing confidential data. According to OpenAI, the feature does not stop prompt injections from appearing in processed content but focuses on blocking unauthorised transfer of data, which the company identifies as a critical stage in such attacks.
With Lockdown Mode enabled, live web browsing is limited to cached content only, while Deep Research and Agent Mode become unavailable. The chatbot can no longer automatically download files for analysis, though manual uploads remain functional, and search results may appear incomplete or outdated.
Apps, connectors, and external integrations face limitations as well: live access and write actions are blocked, though synced experiences continue for personal and self-serve business accounts. Shopping-agent features and financial integrations are also disabled during this time.
The company noted that Lockdown Mode does not impact memory, conversation sharing, image generation, or the use of chats for improving its AI models. It also leaves network access within Codex unaffected. However, the feature cannot be used alongside Developer Mode, as enabling one automatically disables the other.
Essential Business Intelligence, Continuous LIVE TV, Sharp Market Insights, Practical Personal Finance Advice and Latest Stories — On NDTV Profit.