OpenAI is turning on Lockdown mode for ChatGPT users on June 12, 2026.
Lockdown mode limits the model’s ability to fetch live web content or retrieve images, effectively cutting off the features that prompt‑injection attacks exploit. The setting is being rolled out to Enterprise accounts and to Plus‑tier subscribers; free users will not have access.
The change matters because it offers a concrete defense against a known vector of prompt manipulation, something many developers have struggled to mitigate with code alone. The trade‑off is a loss of real‑time data, meaning the model can no longer pull in fresh information from the internet.
In practice, you get a safer assistant at the cost of up‑to‑date answers – a typical security versus convenience bargain.