The card discloses that Meta conducted recurring red teaming exercises and targeted evaluations across three critical risk categories, CBRNE, child safety, and cyber attack enablement, and describes the methodology and scope of those evaluations.
This analysis describes what Meta's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision documents the scope and methodology of Meta's pre-release safety evaluation for Llama 4, providing institutional deployers with a basis for assessing the categories of risk that were formally evaluated prior to release and those that were not.
The document states that safety evaluations covered CBRNE-related prompts, child safety outputs, and cyberattack enablement capabilities, and that these evaluations informed safety fine-tuning. The card also notes that testing could not cover all scenarios and that developers should conduct additional testing for their specific applications.
Cross-platform context
See how other platforms handle Red Teaming and Critical Risk Evaluation and similar clauses.
Compare across platforms →"We conduct recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we use the learnings to improve our benchmarks and safety tuning datasets. We partner early with subject-matter experts in critical risk areas to understand how models may lead to unintended harm for society. We spend additional focus on the following critical risk areas: CBRNE (Chemical, Biological, Radiological, Nuclear, and Explosive materials) helpfulness... Child Safety... Cyber attack enablement.Excerpt from Meta's Llama 4 Model Card
(1) REGULATORY LANDSCAPE: The disclosure of CBRNE and child safety evaluations engages potential obligations under the EU AI Act for transparency about safety testing methodologies for general-purpose AI models.
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
This provision documents the scope and methodology of Meta's pre-release safety evaluation for Llama 4, providing institutional deployers with a basis for assessing the categories of risk that were formally evaluated prior to release and those that were not.
The document states that safety evaluations covered CBRNE-related prompts, child safety outputs, and cyberattack enablement capabilities, and that these evaluations informed safety fine-tuning. The card also notes that testing could not cover all scenarios and that developers should conduct additional testing for their specific applications.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Meta.