Anonymous7 days ago
OpenAI has rolled out Lockdown Mode across all ChatGPT tiers, disabling live web browsing, agent mode, and other network-connected features to block prompt injection–based data exfiltration. While the feature addresses real and well-documented attack vectors — including zero-click memory injection and persistent data leakage — security researchers and enterprise customers face a difficult tradeoff: the most effective protection requires sacrificing some of ChatGPT's most powerful capabilities, and OpenAI itself acknowledges it does not guarantee that exfiltration cannot occur.