OpenAI、プロンプトインジェクションによるデータ流出を防ぐ「Lockdown Mode」を公開
OpenAIはChatGPT向けにLockdown Modeを公開しました。Webアクセスとエージェント機能を無効化してプロンプトインジェクション攻撃を防ぎ、エンタープライズユーザーに機密データを扱うより安全な環境を提供します。
openai introduced lockdown mode, a new security setting for chatgpt that disables web access, deep research, agent mode, and file downloads to prevent data exfiltration via prompt injection. the feature restricts all connections to external services, effectively blocking attackers from manipulating the model to send data to the open internet. it targets enterprise users and individuals handling sensitive information, allowing temporary toggles for specific workflows. this release highlights the ongoing industry challenge of prompt injection, which remains unsolved despite years of research. admins in managed workspaces can enforce the mode via rbac, while personal users can activate it in security settings.