OpenAI introduces Lockdown Mode and Elevated Risk labels for ChatGPT to combat data exfiltration and prompt injection attacks.
OpenAI has launched Lockdown Mode and Elevated Risk labels for ChatGPT, aiming to mitigate data exfiltration risks from prompt injection attacks. Lockdown Mode is an optional advanced security setting that restricts ChatGPT's external connectivity by disabling tools like live web browsing (restricting it to cached content), Deep Research, Agent Mode, Canvas networking, and web-based image/file downloads. Alongside this, Elevated Risk labels serve as a visual UI warning, notifying users when they are engaging with features that carry a higher security exposure.
While Lockdown Mode represents a crucial step forward for enterprise data security, disabling key capabilities like live web browsing and Deep Research highlights a fundamental trade-off: current AI architectures cannot be made secure without crippling their most advanced agentic features.
* By shifting the defense mechanism to blocking data exfiltration channels rather than preventing prompt injections, OpenAI pragmatically accepts that prompt injection remains an unsolved model-level vulnerability.
* Elevated Risk labels act as a lightweight, low-friction compliance mechanism that transfers security awareness back to the user without degrading the default experience.
* Enterprise admin enforcement settings are a major win, providing IT departments with the granular controls needed to approve ChatGPT usage in highly regulated industries.
DISCOVERED
53d ago
2026-06-08
PUBLISHED
53d ago
2026-06-08
RELEVANCE
AUTHOR
OpenAIDevs
