OpenAI has introduced a new security feature called Lockdown Mode, now available to eligible Free, Go, Plus, Pro, and self-serve ChatGPT Business accounts. The feature aims to protect users handling sensitive information by reducing risks related to prompt-injection attacks, which can lead to data theft. Lockdown Mode restricts several web-connected capabilities to enhance security, the company said this week, according to medianama.com.
Lockdown Mode disables or limits features such as Deep Research, Agent Mode, and live web browsing, which connect AI models to external web content. Prompt injection involves attackers embedding malicious instructions within content like webpages or documents, causing AI models to reveal sensitive data or perform unintended actions. OpenAI noted that while Lockdown Mode substantially reduces these risks, it does not guarantee complete protection, as vulnerabilities may persist through third-party apps, cached data, or new attack methods, medianama.com reported.
The rollout addresses growing concerns about data exfiltration through AI prompt manipulation, a technique increasingly exploited to bypass safeguards. By limiting AI access to external web data, Lockdown Mode seeks to prevent attackers from injecting harmful prompts that could compromise confidential information. This move follows broader industry efforts to enhance AI security amid rising adoption in sensitive sectors. OpenAI’s approach aligns with similar measures by other AI providers aiming to balance functionality with data protection, medianama.com noted.
Lockdown Mode is now available as an optional setting for qualifying ChatGPT users, reflecting OpenAI’s commitment to evolving AI security. The company’s FAQ clarifies that while the feature reduces prompt injection risks, users should remain vigilant about third-party app permissions and data handling practices. OpenAI’s announcement this week marks a significant step in securing AI interactions against emerging cyber threats.