OpenAI has introduced two new security measures for ChatGPT users: lockdown mode and “elevated risk” flags. These two security measures work in tandem to give users better information about risks and more control over the system.
Lockdown mode is an advanced security feature that is optional and meant for a small group of users who might be at a higher risk of cybersecurity attacks. This includes executives, security professionals, and teams at major organisations. It is not meant for everyday users, but rather for those who need an added layer of defence.
When enabled, lockdown mode tightly limits how ChatGPT interacts with external systems. It deterministically disables certain tools and features that attackers might attempt to exploit through prompt injection. The goal is to prevent sensitive data from being extracted through hidden or malicious instructions.
For example, web browsing in lockdown mode is restricted to cached content. That means no live network requests leave OpenAI’s secure network environment. If the system cannot guarantee strong data protection for a feature, that feature may be turned off entirely.
Business plans already include enterprise-grade security protections. Lockdown mode adds to these with more restrictive controls. Workspace administrators can turn on lockdown mode by creating a dedicated role in workspace settings. When enabled, it imposes further restrictions on top of existing security settings.
Administrators still have flexibility. They can choose which apps and even which actions on those apps are available while lockdown mode is on. There are also separate tools for compliance logging that give in-depth information on app use, shared data, and connected systems.
Lockdown mode is available now for enterprise and sector-specific plans, with consumer plans to follow in the coming months.
Story continues below this ad
Better warnings for higher-risk Features
The second update is about making things clearer. There are some AI features that access the web or other systems that could pose risks that are still developing in the industry. While most users may be willing to take this risk for the benefits, others might want to err on the side of caution, particularly when working with sensitive data.
To better inform users, certain features in ChatGPT, ChatGPT Atlas, and Codex will have a consistent “elevated risk” warning.
These categories clearly define what is affected by turning on a feature, what risks could occur, and when it is necessary to use it. For example, the ability to access a network through coding utilities could raise the risk of vulnerabilities, and this category will clearly state this.
Security features will continue to develop with new threats. As the security features become better and the risks are mitigated, the “elevated risk” category could be eliminated from certain features. Certain new features might also necessitate the category.
Story continues below this ad
Lockdown mode and better risk communication mean that the goal is simple: to better protect users and provide them with clearer options as artificial intelligence is increasingly integrated into their work.
