Archive — US — techcrunch
OpenAI unveils Lockdown Mode to protect sensitive data from prompt injectionattacks
OpenAI launched Lockdown Mode to enhance protection against prompt injection attacks that could compromise sensitive data. This mode disables live web browsing and image retrieval, focusing on security for users managing sensitive information, though the company warns it does not eliminate all risks. Lockdown Mode is available for self-serve ChatGPT Business accounts and select personal accounts.
Key points
- OpenAI introduced Lockdown Mode to protect against prompt injection attacks.
- Prompt injection attacks involve inserting malicious instructions into chatbots via web content.
- Lockdown Mode disables live web browsing and image retrieval, limiting access to cached content.
- The feature is primarily for users and organizations managing sensitive data and seeking robust protection.
- Despite its implementation, vulnerabilities to prompt injections remain, especially from cached content or uploaded files.
Why it matters
- The rise of AI-driven content generation has increased the risk of prompt injection attacks.
- Prompt injection threats have been a focal point in the cybersecurity landscape as AI tools proliferate.
- As of 2025, organizations are increasingly concerned with data security, making features like Lockdown Mode essential.
Lockdown Mode showcases OpenAI's commitment to enhancing security for sensitive data handling, reflecting a broader trend in 2025 towards integrating robust data protection measures within AI technologies.