New ChatGPT Lockdown Mode to Mitigate Prompt Injection and Data Exfiltration Attacks

OpenAI announced ChatGPT Lockdown Mode, a security feature that restricts outbound network functionality (e.g., live web browsing, image retrieval, agent mode, canvas networking, deep research, and external file downloads) to reduce the risk of data exfiltration from prompt-injection attacks. The announcement explains risk tiers for apps/connectors, administrator deployment and RBAC requirements, limitations of the feature (it does not stop injections or guarantee complete protection), and available audit/logging options for managed workspaces.