OpenAI has introduced a new feature called Lockdown Mode to protect against prompt injection attacks. While it sounds like an impenetrable shield, users can still generate images and share cached content. The mode disables live web browsing and deep research, ensuring that only previously saved data is accessed.

The company emphasises that even with Lockdown Mode active, ChatGPT remains vulnerable to malicious instructions hidden in old content or uploaded files. This means that sensitive information could still slip through the cracks, albeit with reduced frequency.

Lockdown Mode is targeted at businesses and individuals handling sensitive data. OpenAI’s rollout strategy involves offering it first to self-serve business accounts and eligible personal ones, positioning it as a crucial tool for organisations dealing with high-risk content.

The feature highlights the ongoing battle between AI innovation and security. While Lockdown Mode represents a step forward in protecting user data from external threats, the reality is that no system can be 100% foolproof. It’s an interesting milestone on the path to safer AI interactions online.