ChatGPT: OpenAI Launches Lock Mode for Enhanced Security

⚡

Key Takeaways

1OpenAI introduces a lock mode on ChatGPT to protect against query infiltration.

2This mode reduces the AI's access to the web and external services, enhancing security.

3The lock mode is intended for users handling sensitive data, but it is not mandatory.

💡Why it matters — Data security is becoming crucial as attacks on AI systems become more sophisticated.

OpenAI Enhances ChatGPT Security with Lock Mode

OpenAI has recently introduced a lock mode for ChatGPT, providing users with an additional layer of security. This feature aims to protect accounts against an attack technique known as query infiltration, which allows hackers to inject malicious commands into the AI to alter its behavior. OpenAI already enables the securing of ChatGPT accounts through an authentication app, but this is not always sufficient to prevent leaks of sensitive data.

Understanding Query Infiltration

Query infiltration is a method by which malicious commands are hidden in online texts. For example, a hacker might post a malicious prompt on a forum, prompting large language models (LLMs) to redirect users to phishing sites. When a user employs an LLM to read and summarize the discussion on the forum, the generated summary could inadvertently invite the user to visit the hacker's page without their awareness.

Reduced Features for Increased Security

To counter this type of attack, ChatGPT's lock mode limits the AI's access to the web and external services. This means that the artificial intelligence will be less useful but more secure. This mode is available for all types of accounts, although OpenAI specifies that it is not designed for everyone. "It is intended for individuals and organizations that handle sensitive data and seek stricter protection against data exfiltration risks related to rapid injection," explains the AI lab.

Activating Lock Mode on ChatGPT

To activate this mode, interested users must go to the web version of ChatGPT, click on their name at the bottom of the left sidebar, and then select "Settings." In the "Security" tab, they can enable "Lock Mode." This mode is optional, and OpenAI emphasizes that it is not meant for all users.

Responding to Emerging Threats

Although query infiltration is not yet considered a major risk, OpenAI acknowledges that its impact could grow as attackers develop more sophisticated methods. This new security feature illustrates the increasing challenges associated with using AIs on the web, as users seek ways to protect their sensitive data.