OpenAI has started deploying a new security feature called Lockdown Mode for ChatGPT, aimed at reducing the attack surface for data exfiltration through prompt injection exploits. The rollout, which targets eligible personal accounts, restricts certain tool functionalities that could be weaponised by adversaries to siphon sensitive information.
What Lockdown Mode Does
The feature is designed with people and organisations that routinely handle sensitive data in mind. Lockdown Mode limits access to specific ChatGPT tools and capabilities that might otherwise be abused during a prompt injection attack — a technique where malicious instructions are embedded within content the AI processes, potentially causing it to perform unintended actions such as transmitting data to external endpoints.
The mode is available to eligible logged-in users across all subscription tiers. By making the feature broadly accessible rather than gating it behind a premium paywall, OpenAI is signalling that robust security controls should be treated as a baseline expectation, not a luxury add-on.
Why It Matters for Security-Conscious Users
Prompt injection has emerged as one of the most persistent and difficult-to-mitigate vulnerability classes in large language model (LLM) deployments. Attackers can craft adversarial inputs that trick the model into executing hidden instructions — for example, reading private documents and sending their contents to a third-party server via a connected tool or plugin.
Lockdown Mode offers a practical, defence-in-depth option for IT and security teams that want to reduce risk without completely disconnecting from AI-powered tools. For organisations evaluating how to safely integrate ChatGPT into workflows involving confidential data, the feature provides an additional layer of assurance.
That said, it is important to set realistic expectations. No single feature can provide airtight protection against all sophisticated, targeted prompt injection attacks. Lockdown Mode narrows the exploitation surface, but users and security teams should continue to apply broader security hygiene practices, including careful review of connected integrations and monitoring for anomalous behaviour.
A Broader Industry Signal
OpenAI's decision to roll out Lockdown Mode across all tiers is a noteworthy strategic move. It frames AI security not as a premium differentiator but as a fundamental safeguard that every user should have access to. Whether competing AI providers follow suit with similar built-in, user-configurable security modes remains an open question — but the precedent could accelerate industry-wide adoption of such features.
Security researchers and IT professionals will be watching closely for real-world assessments of the feature's effectiveness against advanced prompt injection techniques. As with any new security control, its true value will be measured over time through independent testing, incident reports, and the evolving threat landscape.
For now, users who routinely process sensitive or regulated data through ChatGPT have a new tool worth exploring. Enabling Lockdown Mode is a low-friction step that can meaningfully reduce exposure while the broader AI security ecosystem continues to mature.
OpenAI 已開始為 ChatGPT 部署一項名為「限制模式」的新安全功能，旨在透過減少提示注入漏洞的攻擊面，來降低數據竊取風險。這項針對符合條件的個人帳戶推出的更新，限制了某些可能被對手利用來竊取敏感資訊的工具功能。
限制模式的功能
這項功能的設計考量了經常處理敏感數據的個人和組織。限制模式限制了對特定 ChatGPT 工具和功能的存取，這些功能在提示注入攻擊期間可能被濫用——提示注入是一種將惡意指令嵌入 AI 處理內容中的技術，可能導致 AI 執行非預期的操作，例如將數據傳輸至外部端點。
所有訂閱級別的符合條件的登入用戶均可使用此模式。OpenAI 透過廣泛提供此功能，而非將其置於付費高級方案之後，表明穩健的安全控制應被視為基本期望，而非奢侈附加功能。
對注重安全的用戶為何重要
提示注入已成為大型語言模型（LLM）部署中最持久且最難緩解的漏洞類別之一。攻擊者可以設計對抗性輸入來欺騙模型執行隱藏指令——例如讀取私人文件，並透過連接的工具或 plugin 將其內容傳送到第三方伺服器。
對於希望在不完全斷開與 AI 驅動工具連接的情況下降低風險的 IT 和安全團隊而言，限制模式提供了一個實用的縱深防禦選項。對於正在評估如何在涉及機密數據的工作流程中安全整合 ChatGPT 的組織，此功能提供了額外的保障層。
話雖如此，設定現實的期望很重要。沒有任何單一功能能夠對所有複雜的、有針對性的提示注入攻擊提供滴水不漏的保護。限制模式縮小了可被利用的範圍，但用戶和安全團隊應繼續遵循良好的安全習慣，包括仔細審查連接的整合以及監控異常行為。
來自行業的更廣泛信號
OpenAI 決定在所有訂閱級別推出限制模式是一項值得注意的戰略舉措。它將 AI 安全定位為每位用戶都應獲得的基本保障，而非高級差異化優勢。其他競爭性 AI 提供商是否會跟進，推出類似的內建、用戶可配置安全模式，仍有待觀察——但此先例可能會加速整個行業採用此類功能。
安全研究人員和 IT 專業人士將密切關注此功能對抗高級提示注入技術的實際效果評估。與任何新的安全控制措施一樣，其真正價值將透過獨立測試、事件報告和不斷演變的威脅格局，隨時間推移來衡量。
目前，經常透過 ChatGPT 處理敏感或受監管數據的用戶有了一個值得探索的新工具。啟用限制模式是一個低門檻的步驟，在更廣泛的 AI 安全生態系統持續成熟的同時，可以有意義地減少風險敞口。