OpenAI is rolling out Lockdown Mode for ChatGPT, a security setting that mitigates prompt injection-based data exfiltration by disabling live web browsing, agent mode, deep research, image retrieval, Canvas networking, and file downloads. The feature is available across all plans including Free and Plus. It does not prevent prompt injections from occurring — malicious payloads in cached pages or uploaded files can still influence the model — but it eliminates the outbound channels attackers use to steal data. OpenAI explicitly acknowledges this is not a complete fix, and residual risk remains through enabled apps or unforeseen capability combinations. The trade-off is substantial: most of ChatGPT's agentic and research capabilities are disabled when Lockdown Mode is active. The launch comes amid growing concern over prompt injection attacks targeting AI agents from Anthropic, Google, and Microsoft, with researchers demonstrating hijacks via GitHub Actions integrations.
