OpenAI introduced Lockdown Mode on June 5, 2026 as an optional setting for a small cohort of users. The feature tightens input handling to block prompt‑injection techniques that attempt to hijack the model’s responses. It is currently available only to select accounts that have been invited to opt in via the user dashboard.
The addition matters because prompt injection remains a practical attack vector on large language models, and OpenAI’s default safeguards have proven insufficient for high‑risk deployments. By giving administrators a stricter mode, the company offers a concrete tool for those who need stronger guarantees without overhauling the entire service.
Lockdown Mode is not a blanket solution; it targets a niche audience and will likely stay behind a paywall, leaving most developers to rely on existing precautions.