Llama 3: CyberSec Eval 2 Unveiled
In a concerted effort to advance the development of Llama 3 in a responsible manner, a suite of resources has been crafted to facilitate its ethical utilization by others. Central to this endeavor are newly introduced trust and safety tools, prominently featuring CyberSec Eval 2. This upgraded evaluation framework builds upon its precursor, extending its scope to include assessments on susceptibility to prompt injection, automated offensive cybersecurity capabilities, and propensity to exploit code interpreters. These enhancements complement the existing evaluations pertaining to insecure Coding Practices and Cyber Attack Helpfulness, reinforcing a commitment to fostering secure and ethical AI ecosystems.