Meta AI has developed CYBERSECEVAL 3, a tool designed primarily for evaluating cybersecurity risks and benefits related to AI systems. The tool focuses on LLMs such as the Llama 3, providing a comprehensive risk assessment.

The CYBERSECEVAL 3 tool carries out a series of empirical tests to scrutinize the cybersecurity implications of AI systems. Its functionalities include automating social engineering, assisting in manual offensive cyber operations, and enabling autonomous cyber operations. It also includes autonomous software vulnerability discovery and exploitation.

Meta AI claims that the key advantages of CYBERSECEVAL 3 are that it provides a complete evaluation of cybersecurity risks, allows for testing of LLMs' potential to automate cybersecurity attacks, and is capable of autonomous operation. This helps manage the risks well and reinforces security in the AI environment.

Analyst QuickTake : CYBERSECEVAL 3 builds on the previous evaluations, CYBERSECEVAL 1 and 2, which examined LLM risks, such as exploit generation and insecure code. This update aligns with Meta's commitment to responsible AI development, shown in their release of the Llama 3 models. For instance, alongside Llama 3.1 , Meta introduced two key safety tools: Llama Guard and Prompt Guard.