OpenAI Introduces Lockdown Mode for Enhanced ChatGPT Security
OpenAI has launched Lockdown Mode, a new security feature designed to reduce prompt injection risks and protect sensitive information in ChatGPT.
SociapaHQ
OpenAI has begun rolling out Lockdown Mode, a new optional security feature for users and organisations that handle sensitive information and need additional protections against data exposure through prompt injection.
We are rolling out the feature to eligible personal accounts, including Free, Go, Plus and Pro users, as well as self-serve ChatGPT Business accounts. Users who don’t see the option in settings yet may get it at a later stage, OpenAI said, adding that access will be expanded gradually.
Lockdown Mode is intended to reduce the risk of data exfiltration where an attacker may use prompt injection attacks to extract sensitive information, the company says. To do this, the feature limits the number of tools and capabilities that require internet connectivity or external services.
Also Read: What Does The UK’s Google AI Ruling Mean For Publishers?
Security Restrictions:
When in Lockdown Mode, ChatGPT’s web browsing is limited to cached content only. Search results may therefore be unavailable, incomplete, or out of date. The environment also disables Deep Research, Agent Mode, file downloads for data analysis, and the ability to approve Canvas-generated code that needs network access.
According to reports, OpenAI said it also impacted image-related functionality. Users will still be able to upload images and generate new ones where supported, but the chatbot may not display images in standard responses or retrieve them from the web.
The feature is intended to block outbound network requests that could be used to transfer sensitive data outside of a secure environment, the company said.
Also Read: Meta Oversight Board Criticises Account Ban Process, Calls for Greater Transparency
However, OpenAI noted that Lockdown Mode does not offer bulletproof protection against prompt injection attacks. Even if the feature is enabled, malicious instructions embedded in cached webpages or uploaded files can still affect the behaviour or accuracy of ChatGPT responses.
The company also said that Lockdown Mode does not alter settings around memory, file uploads, sharing of conversations or the use of conversations to inform AI model training. Those controls remain under separate user settings and organisational policies.
Personal account holders and self-serve ChatGPT Business subscribers can find Lockdown Mode in the Security section of Settings.
OpenAI added that Lockdown Mode cannot be used simultaneously with Developer Mode. Enabling one feature will automatically disable the other.
