ChatGPT Introduces Lockdown Mode
To Reduce Prompt Injection Risks and Protect Sensitive Data
News Desk
08 June 2026, 15:50
OpenAI has introduced a new security feature called Lockdown Mode for ChatGPT, aimed at reducing the risk of sensitive information being exposed through prompt injection attacks.
The feature is designed for organisations and individuals who work with confidential data and require additional safeguards when using AI tools that interact with external content. According to OpenAI, Lockdown Mode limits several ChatGPT capabilities that could potentially be exploited by attackers to extract sensitive information.
Prompt injection attacks occur when malicious instructions are hidden within webpages, documents, or other content processed by an AI system. These instructions can influence the model's behaviour in unintended ways, potentially leading to security and privacy risks.
What Lockdown Mode Changes
When Lockdown Mode is enabled, ChatGPT disables a number of features that rely on accessing external sources. These include live web browsing, viewing images retrieved from the internet, deep research tools, and agent-style functionality that can perform tasks across multiple systems.
Users can still generate images within ChatGPT, but the model will be restricted to cached web content instead of accessing information from the live internet.
According to OpenAI, these restrictions are intended to reduce opportunities for attackers to use outbound network requests as a channel for data exfiltration.
OpenAI Says Risks Are Not Completely Eliminated
OpenAI has cautioned that Lockdown Mode is not a complete solution to prompt injection threats.
In a recent blog post, the company explained that the feature is designed to prevent the final stage of data exfiltration by limiting network requests that could transfer sensitive information to an attacker. However, it does not stop prompt injections from appearing in content that ChatGPT processes.
The company noted that malicious instructions could still be present in cached web content or uploaded files, meaning they may continue to affect the model's responses or accuracy in some situations.
Availability and Compatibility
Lockdown Mode is currently being rolled out to self-serve ChatGPT Business customers as well as eligible individual users.
The feature can be enabled through the Security section of ChatGPT's settings menu under Advanced Security.
OpenAI also confirmed that Lockdown Mode cannot be used alongside Developer Mode. Activating one feature will automatically disable the other.
Growing Focus on AI Security
The launch of Lockdown Mode reflects growing industry efforts to address security challenges associated with advanced AI systems. As AI tools gain access to web content, enterprise data, and third-party services, prompt injection attacks have emerged as one of the key concerns for organisations deploying AI in business environments.
By introducing stricter controls over external access, OpenAI is offering users an additional layer of protection while acknowledging that broader security risks associated with prompt injections remain an ongoing challenge for the industry.
