Llama 3.3 Use Policy

Llama 3.3 Acceptable Use Policy is a governance document published by Meta AI, dated 2026-04-08. It binds any person or entity who accesses or uses Llama 3.3, requiring agreement as a condition of use. The policy's stated purpose is promoting "safe and fair use" of the model. This document is a use policy, not a model card; it provides no technical specifications, evaluation results, or safety research findings.

Not disclosed in this document.

Not disclosed in this document.

Not disclosed in this document.

The policy enumerates four top-level categories of prohibited use. The first bars violations of law or others' rights, with subcategories covering violence and terrorism; child exploitation including CSAM; human trafficking and sexual violence; illegal distribution of materials to minors without required age-gating; sexual solicitation; unauthorized professional practice; unlicensed collection or inference of private or sensitive personal information; IP infringement; malicious code creation; and intentionally circumventing "usage restrictions or other safety measures, or to enable functionality disabled by Meta." The second category bars facilitating activities presenting a risk of death or bodily harm, including military and ITAR-regulated uses, weapon development, illegal drugs, critical infrastructure operation, and self-harm or harm-incitement content. The third bars intentional deception, including fraud, disinformation, defamation, spam, non-consensual impersonation, misrepresenting AI outputs as human-generated, and fake online engagement. The fourth requires users to disclose known dangers to end users and prohibits interaction with third-party tools designed to generate unlawful content when representing their outputs as associated with Meta or Llama 3.3.

Access is governed by the Llama 3.3 Community License Agreement, and the policy is maintained at llama.com/llama3_3/use-policy. Individuals domiciled in, or companies with a principal place of business in, the European Union are excluded from rights granted under Section 1(a) of that agreement for any multimodal models included in Llama 3.3; this restriction does not apply to end users of a product or service that incorporates those multimodal models. Meta provides four reporting channels: GitHub issues for model problems, a Facebook developer feedback portal for risky model outputs, Facebook's whitehat security program for bugs and security concerns, and LlamaUseReport@meta.com for policy violations or unlicensed uses.

Not disclosed in this document.

The document does not include version deltas, changelog entries, or comparisons to any prior Llama use policy.