Is OpenAI’s New Lockdown Mode an Admission That Default ChatGPT Was Never Safe Enough?

OpenAI introduced Lockdown Mode to reduce the risk of prompt injection–driven data exfiltration by disabling or limiting features that provide outbound network paths (live web access, Agent Mode, file downloads, image-url channels, etc.). The article explains that Lockdown Mode is a last-line defense — it blocks exfiltration channels but does not prevent injected instructions from reaching the model — and is intended for users with elevated risk profiles who prioritize confidentiality over some functionality.