
OpenAI's Lockdown Mode disables web access, Deep Research, Agent Mode, file downloads, and web image display to prevent prompt injection attacks. The feature blocks network exfiltration attempts but does not stop manipulative prompts from influencing model responses. Users handling sensitive data can disable internet-connected features to reduce data exposure. Lockdown Mode limits live search to cached results, which may be outdated or missing. OpenAI states prompt injection is not currently a major risk but could grow as attack methods evolve.
Tap to vote and see what everyone thinks.
OpenAI launches Lockdown Mode for ChatGPT security
Summary by ByteBrief