OpenAI mentions that Lockdown Mode does not completely eliminate the risk of dangerous instructions or prompt injection.

This is an ingested summary. Full reporting is available at the primary source below.