OpenAI has released Lockdown Mode which is a potent and latest security capability for AI models. The target is to ramp down the possibility of data exfiltration by malicious prompt injection attacks. The feature is currently on offer to personal accounts, self-serve ChatGPT Business clients as well as managed enterprise spaces. Prompt injection attacks involve malicious steps being put inside content that is processed by AI application(s). Lockdown Mode seeks to halt the ending stage of a prompt injection attack. In that stage unapproved transfer of critical data to the attacker’s systems is carried out using outbound network requests.

Lockdown Mode puts restrictions on certain tools as well as capabilities in OpenAI products that have the capability to connect to the web and/or external services. The aim is not to halt the occurrence of prompt injections. Instead, the objective is to curtail potential routes by which critical data could be exfiltrated. Hence, Lockdown Mode disables certain features. We take a look at the affected features. Users can’t browse the web in real-time. The former are limited to access just cached content. Images cannot be retrieved from the web or displayed on demand from the web. Thankfully, images can still be generated by AI tools. Deep Research and Agent mode are fully disabled. Users cannot give permission to Canvas-generated code to access external networks. Finally, ChatGPT does not have permission to download files for the purpose of data analysis. Users have permission to only manually upload files.

Note that Lockdown Mode and Developer Mode are mutually exclusive. When one is enabled the other is disabled and vice versa. Lockdown Mode places certain instructions on the functioning of apps, connectors as well external integrations. Live connector access and write actions are not permitted. For your knowledge, shopping-agent features as well as financial integrations will not be available in Lockdown Mode. Codex will continue have network access even in Lockdown Mode.

One can enable Lockdown Mode by navigating to the Settings’ Security section. As per OpenAI, the Lockdown Mode cannot get rid of all the prompt injection risks completely. Malicious content contained in certain uploaded files as well as in cached content can trick the AI model to submit incorrect answers.

SOURCES:-

https://cybersecuritynews.com/chatgpt-lockdown-mode/

https://thehackernews.com/2026/06/new-chatgpt-lockdown-mode-limits-tools.html

https://techcrunch.com/2026/06/06/openai-unveils-lockdown-mode-to-protect-sensitive-data-from-prompt-injection-attacks/

https://www.gadgets360.com/ai/news/chatgpt-lockdown-mode-prompt-injection-attacks-protection-data-theft-risk-openai-11605307/amp

https://pulse2.com/openai-lockdown-mode/