Meta AI✓

Stav AI Act assessment

Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.

• transparencyUnknown

  Llama 3.x series includes model cards, system cards, and partial training data documentation published on HuggingFace and llama.com — above-average transparency for open-weights models.

• communityUnknown

  Meta AI Research (FAIR) is a prolific and well-regarded academic AI research publisher, with substantial contributions to the open research community through papers, datasets, and the Llama open-weights model family.

• securityUnknown

  Meta operates one of the largest and most established bug bounty programmes in the industry, with a long track record of paying out significant awards for responsible vulnerability disclosures.

• transparencyUnknown

  Muse Spark Safety & Preparedness Report published alongside model launch, including third-party evaluation by Apollo Research. This represents a positive step toward frontier model safety transparency.

• stabilityUnknown

  Meta is one of the world's largest technology companies with a well-capitalised parent corporation, Hyperion data centre investments, and a clear multi-year AI roadmap. The risk of model abandonment or organisational collapse is negligible.

• complianceUnknown

  Llama 3.3 and earlier non-multimodal Llama models are EU-compatible (no licence exclusion) and can be self-hosted by EU enterprises, providing full data sovereignty when deployed on EU infrastructure.

Published safeguards & certifications

Flagged issues

2

• Meta has the worst GDPR enforcement track record of any major AI model creator: the 2023 €1.2B fine (EU-US data transfers) and ongoing disputes with the Irish DPC over EU user data being used for AI training constitute a material risk signal for EU regulated-sector customers.high
• Meta uses EU social media user data (Facebook/Instagram public posts) for AI model training under an opt-out mechanism that privacy organisation noyb deemed insufficiently transparent, filing 11 regulatory complaints. This affects the training data provenance of future Llama and Muse models.medium

Privacy policy review

Country

🇺🇸 United States

Open-weights policy

Mixed

Models in catalogue

0

Website

https://ai.meta.com

HuggingFace

meta-llama · Verified · 79,995 followers

HF published

70 models · 11 datasets · 4 papers

Creator profile

Meta AI is the AI research and product division of Meta Platforms, a US-incorporated public company headquartered in Menlo Park, California. It is the originator of the Llama model family — the most widely deployed open-weights large language models in the world — released under custom community licences that are not OSI-certified open source. For EU regulated enterprises, Meta presents a layered compliance risk: it holds the record GDPR fine (€1.2 billion for unlawful EU–US data transfers), has received multiple subsequent DPC and DMA enforcement actions, publicly refused to sign the voluntary EU GPAI Code of Practice, and has had its flagship Llama models classified by regulators as GPAI models with systemic risk. The lab is actively investing in AI and remains operationally strong, but its adversarial relationship with EU regulators and persistent data-sovereignty concerns make it a high-scrutiny supplier for European regulated-sector customers.

Stav editorial summary