OpenAI has introduced a new security setting for ChatGPT users who want tighter controls when working with confidential or sensitive information.
The feature, known as Lockdown Mode, is intended to reduce the risk posed by prompt injection attacks. These attacks can happen when hidden instructions are embedded in webpages, uploaded files, or other external sources that an AI chatbot is asked to process.
In some cases, the concealed commands may influence a response or attempt to expose information that should remain private.
Once the setting is activated, ChatGPT will no longer be able to browse the live web. Instead, it will rely on cached online material stored within OpenAI’s network. The retrieval and display of images from the internet will also be restricted, although users can still create AI-generated images.
Lockdown Mode also switches off deep research and agent mode, two tools that allow ChatGPT to carry out more complex tasks and interact with a wider range of sources. The trade-off is that users will have fewer features available, but the chatbot will have fewer opportunities to encounter harmful instructions or unintentionally share sensitive data.
OpenAI has made clear that the setting is not a complete defence against prompt injection. Malicious instructions may still “appear in cached web content or in an uploaded file, and could still affect the behavior or accuracy of a response.”
The company said the additional safeguards are aimed at people and businesses that place a higher priority on data protection than access to the platform’s full range of capabilities.
“Lockdown Mode is not intended for everyone,” OpenAI says. “It is designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks related to prompt injection.”
The feature is being made available to self-serve ChatGPT Business customers and eligible personal accounts. It is likely to be particularly useful for professionals and organisations that use AI tools while handling private client records, internal documents, or commercially sensitive material.