OpenAI rolls out Lockdown Mode to curb prompt injection

OpenAI Lockdown – OpenAI has introduced Lockdown Mode, a new ChatGPT feature aimed at reducing data exfiltration risks from prompt injection attacks hidden in web pages or other content sources. The company says it limits several browsing and research capabilities, but it also
For people who work with sensitive data, “prompt injection” isn’t a theoretical threat—it’s a practical one. OpenAI’s new Lockdown Mode is built around that reality. designed to reduce the chances that malicious instructions hidden in webpages or other content sources can nudge ChatGPT into leaking or mishandling information.
OpenAI says Lockdown Mode adds stricter protection against prompt injection attacks. The company’s framing is direct: the risk comes when malicious chatbot instructions are buried in content. waiting to be pulled into a conversation. With Lockdown Mode enabled, ChatGPT will disable live web browsing, limiting access to cached content instead. OpenAI also says Lockdown Mode will restrict retrieval and display of images from the web—though the feature still allows image generation.
A few other capabilities are also taken off the table. Lockdown Mode will disable deep research and agent mode. In other words, the feature isn’t just a tweak to how ChatGPT searches—it’s a broader set of guardrails that reduce the pathways through which injected instructions can reach the model.
OpenAI also insists expectations must be realistic. Even with Lockdown Mode turned on, the company says ChatGPT could still be vulnerable to prompt injections. It gives a clear example of how that could happen: prompt injections could “appear in cached web content or in an uploaded file. ” and they could still affect the behavior or accuracy of a response.
So the point isn’t a promise of immunity. It’s risk reduction. OpenAI says Lockdown Mode is “not intended for everyone.” The feature is meant for “people and organizations that handle sensitive data” and want stricter protection from data exfiltration risks tied to prompt injection.
The rollout is also specific. OpenAI says it’s currently rolling Lockdown Mode out to self-serve ChatGPT Business accounts, along with eligible personal accounts.
OpenAI Lockdown Mode ChatGPT Business prompt injection cybersecurity data exfiltration cached content agent mode deep research