Cryptographic Sanctuaries: OpenAI Unveils “Lockdown Mode” to Counter Prompt Injection Risks

This report describes OpenAI's Lockdown Mode as an advanced safety configuration that prevents data exfiltration from prompt-injection attacks by disabling agentic features, blocking automated web/media ingestion, and restricting outgoing network capabilities, while noting usability trade-offs and session management controls.