OpenAI Introduces “Lockdown Mode”, And It’s As Bad as You Might Think

ByMike Hayes

OpenAI just announced a new feature. It’s aimed squarely at companies and people who deal with sensitive information, and it’s been dubbed ‘Lockdown Mode’.

It’s supposed to protect you from prompt injection attacks. This is where someone places malicious prompts either directly into an AI prompt (direct injection) or in a place where AI is likely to be used, such as an email, webpage or PDF file (indirect injection).

While in ‘Lockdown Mode’, ChatGPT can’t do any of the following:

  • browse the web
  • display images in responses (however, you can still upload images and the AI can still generate them)
  • perform “Deep Research”
  • function as an agent
  • download files
  • network the Canvas code generator

As OpenAI puts it:

“Lockdown Mode is not intended for everyone. It is designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks related to prompt injection.”

In other words, you’re probably not important enough.

With AI becoming increasingly integrated into, well, everything (for example, Microsoft Windows leans heavily on Copilot since Windows 10 came out), the dangers of these attacks are becoming more apparent. Business workflows, applications, operating systems–it seems everyone is using AI these days.

While the concept of AI is quite mature, AI assistants that help us perform the mundane tasks of daily office work are still somewhat in their infancy. So, it only seems only natural for developers to want to build in protections for medical professionals, lawyers, and the like so that sensitive patient and client data doesn’t find its way into the hands of people who might want to steal an identity, for example.

Still, it’s more than slightly disconcerting to think that a doctor might dump a patient’s chart or that a lawyer might scan documents that have a client’s social security number on them into an LLM prompt, and it begs the question: why the hell would you put any sensitive data anywhere near an AI prompt in the first place?

Citations

Pearl, Mike “OpenAI Announces Unnerving New ChatGPT Feature Named ‘Lockdown Mode’. https://gizmodo.com/openai-announces-unnerving-new-chatgpt-feature-named-lockdown-mode-2000768425, 2026-06-07

Similar Posts