Loading the catalogue…
Loading the catalogue…
Meta Platforms is a US-incorporated company fully subject to the CLOUD Act and FISA Section 702; its EU obligations run through Meta Ireland Ltd under GDPR, with a long history of major DPC enforcement actions. Llama 3.3 and earlier text-only models remain open-weights and can be self-hosted on EU infrastructure, but Llama 4 multimodal models are contractually excluded for EU-domiciled licensees, Meta has refused to sign the EU GPAI Code of Practice, and its new closed-weights flagship (Muse Spark) is US-only. Combined with a major 2025-2026 AI leadership reshuffle (mass layoffs, Chief AI Scientist Yann LeCun's departure), the picture is one of a well-resourced but strategically unsettled lab whose regulatory posture is openly adversarial toward the EU AI Act.
Llama 4 multimodal models are widely reported to contractually exclude EU-domiciled individuals and companies from licence rights under the Llama 4 Community Licence Agreement/Acceptable Use Policy.
Meta Platforms is a US-incorporated entity fully subject to the CLOUD Act and FISA Section 702; US authorities can compel access to data held globally.
Meta refused to sign the EU GPAI Code of Practice, resulting in a January 2026 EU AI Office 'ecosystem enforcement' investigation into Meta's WhatsApp Business API and continued closer regulatory scrutiny of Llama models.
An EU legal ruling reportedly found Meta's AI training data practices illegal under EU standards with no US federal equivalent, and Meta's opt-out mechanism for training data drew multiple noyb complaints.
Significant AI leadership instability: Chief AI Scientist Yann LeCun departed in November 2025 amid ~600 layoffs from the Superintelligence/FAIR unit in October 2025, and the team that owned Llama development (AGI Foundations) was dissolved after Llama 4's weak reception.
Meta reportedly used an unreleased experimental chat variant of Llama 4 for the LMArena leaderboard submission rather than the model actually shipped, raising disclosure-integrity concerns relevant to regulated-sector due diligence.
Muse Spark, Meta's newest and most capable model, is closed-weights, US-only, and has no published EU availability, licensing, or AI Act compliance documentation as of mid-2026.
Meta's stake in Scale AI (whose former CEO now leads Meta's AI division) creates an indirect connection to Scale AI's US government/defence-sector contract history, worth assessing for supply-chain independence.
Stav’s 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.
The Llama 1–4 model family includes published model cards, technical reports, and open weights (for non-EU-excluded versions), which is comparatively strong disclosure among closed/open-weights hybrid labs.
Llama has surpassed one billion downloads and remains widely adopted across a broad cloud-partner ecosystem (AWS, Azure, Google Cloud, Databricks, and others), reflecting significant developer and enterprise community engagement.
Despite internal reorganisation, Meta continues large-scale, multi-year AI infrastructure investment (e.g. the Hyperion and Prometheus data-centre projects), indicating low near-term risk of abandoning its AI product line.
Llama 3.3 and earlier non-multimodal models remain deployable under a permissive-enough community licence for EU enterprises to self-host on their own EEA infrastructure, preserving a data-sovereignty option even amid Meta's broader EU friction.
Published safeguards & certifications