cyber.hendryadrian.com
Security News · thecyberexpress.com

OpenAI Adds Lockdown Mode and Elevated Risk Labels to ChatGPT

· Original article ↗

Summary

OpenAI introduced Lockdown Mode for high-risk ChatGPT users, restricting network access and some features, plus Elevated Risk labels for network-related capabilities. The company says the controls address prompt-injection risks.

Key points

  • Lockdown Mode is an optional setting aimed at high-risk users and limits web browsing to cached content, according to OpenAI.
  • The mode disables features including images, Deep Research, Agent Mode, network downloads and network access for Canvas-generated code.
  • Workspace administrators can enable the setting for supported enterprise and education plans, control available apps and actions, and review usage through Compliance API Logs Platform.
  • Elevated Risk labels identify network-related capabilities, including Codex network access, and explain their potential risks; OpenAI says the labels may change as mitigations improve.
  • OpenAI says most users do not need Lockdown Mode and plans to make it available to consumers in coming months.

Article Details

Event Type
OpenAI rollout of enterprise security features to mitigate prompt-injection and data-exfiltration risks
Impact
Lockdown Mode applies deterministic restrictions for high-risk users: web browsing is limited to cached content, live external network requests are blocked, and several capabilities—including images, Deep Research, Agent Mode, network downloads, and approval of Canvas-generated code for network access—are disabled. Workspace administrators can configure access and review usage through the Compliance API Logs Platform. Elevated Risk labels warn users when network-related features may increase exposure; they are informational rather than prohibitions. The article reports no specific incident or victim impact.

Vendors

OpenAIOpenAI rolled out Lockdown Mode and Elevated Risk labels to counter prompt injection attacks by adding deterministic infrastructure controls that prevent data exfiltration regardless of prompt manipulation.

Products

ChatGPTThe feature tightly constrains how ChatGPT interacts with external systems through deterministic restrictions that eliminate attack surfaces prompt injection exploits.ChatGPT AtlasThe standardized labeling system appears across ChatGPT, ChatGPT Atlas and Codex whenever users enable network-related capabilities that may increase exposure.ChatGPT EnterpriseWorkspace administrators on ChatGPT Enterprise, Edu, Healthcare and Teachers plans activate Lockdown Mode by creating specialized roles through Workspace Settings.CodexElevated Risk labels mark network-related features (such as Codex network access) to inform users of residual risks and are updated or removed as mitigations improve.Compliance API Logs PlatformWorkspace administrators enable Lockdown Mode via specialized roles and maintain granular app/action controls with visibility through Compliance API Logs Platform.

Related Articles