The Next Web OpenAI has begun rolling out a new security setting called Lockdown Mode across all ChatGPT plans. The feature disables live browsing, agent mode, deep research, image retrieval, Canvas networking and file downloads, aiming to block data exfiltration via prompt‑injection attacks. While it does not prevent malicious payloads from reaching the model, it cuts off outbound pathways that attackers use to steal information. The change arrives as researchers highlight growing risks to AI agents from injection techniques. Lockdown Mode is now available to Free, Go, Plus, Pro and Business users, though OpenAI warns it may limit functionality for many.
Lire la suite