ChatGPT gets new security feature to fight prompt injection attacks

Source: Help Net Security

Author: Anamarija Pogorelec

URL: https://www.helpnetsecurity.com/2026/02/16/chatgpt-lockdown-mode-elevated-risk/

ONE SENTENCE SUMMARY:

OpenAI added ChatGPT Lockdown Mode and Elevated Risk labels to curb prompt injection, restrict tools, and clarify risky integrations enterprise.

MAIN POINTS:

  1. Lockdown Mode is an optional advanced security setting for highly security-conscious users.
  2. Tool access is deterministically constrained to reduce prompt-injection–driven data exfiltration.
  3. Network browsing is limited so no live requests leave OpenAI’s controlled network.
  4. Cached content browsing helps prevent attackers from siphoning sensitive data via the web.
  5. Workspace admins enable Lockdown Mode by creating a dedicated role in settings.
  6. App availability and permitted actions can be selectively configured for Lockdown users.
  7. Current availability includes ChatGPT Enterprise, Edu, Healthcare, and Teachers editions.
  8. Future plans include expanding Lockdown Mode availability to consumer users.
  9. Elevated Risk labels provide in-product guidance for features that increase security exposure.
  10. Labels span ChatGPT, ChatGPT Atlas, and Codex, explaining changes, risks, and appropriateness.

TAKEAWAYS:

  1. Adopt Lockdown Mode to minimize external-system abuse paths during sensitive workflows.
  2. Prefer cached-only browsing when preventing inadvertent data leakage is a priority.
  3. Use role-based controls to enforce stronger security restrictions without disrupting other admin policies.
  4. Treat Elevated Risk labels as decision aids when enabling web/app connectivity capabilities.
  5. Expect risk labeling to evolve and be removed once safeguards sufficiently mitigate threats.