OPENAI GOES ON DEFENSE — LOCKDOWN MODE NOW BLOCKS THE AI HACKS THAT COULD STEAL YOUR SECRETS OpenAI quietly launched a security feature called Lockdown Mode this week that is designed to stop one of the nastiest attack vectors facing AI systems today. The target is prompt injection, the technique where malicious instructions hidden inside web pages, documents, or uploaded files hijack what a chatbot does on your behalf. The threat is not theoretical. As ChatGPT gained access to the web, email, files, and connected apps, the attack surface exploded. A bad actor can embed invisible instructions in a webpage that tell ChatGPT to silently forward your data to an external address. Lockdown Mode cuts off the outbound routes before that can happen. When enabled, the feature disables web browsing, image support, Deep Research, Agent Mode, Canvas networking, live connectors, and file downloads. That is a significant trade-off in functionality, but OpenAI is pitching it specifically at users and businesses handling sensitive data who need tighter controls. There is an honest caveat in OpenAI’s own documentation. The company admits Lockdown Mode does not guarantee your data cannot be exfiltrated. It substantially reduces the risk. That is not the same thing. For now it is the best available option. Lockdown Mode is available to all users including the free tier. Keywords: OpenAI Lockdown Mode, prompt injection attack, ChatGPT security, AI data security