Llama 4
Meta
partial detail 8 method fields published
Evaluation
CBRNE expert-designed evaluations and red teaming
Expert-designed and other targeted evaluations assessing whether Llama 4 could meaningfully increase malicious-actor capabilities to plan or carry out CBRNE attacks, plus additional red teaming and evaluations for content-policy violations in this risk area
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.