- Deployment
- hosted
- Hq Country
- United States
- Hq City
- Menlo Park
- Ownership
- public
- Founded
- 2004
- Training Default
- never
- Residency Options
- US (no EU/regional residency evidenced for the hosted API)
- Gov Access Exposure
- us-cloud-act
- Eu Ai Act Role
- gpai-systemic
- Self Host Option
- Yes
- Retention Default
- unknown — Llama API legal pages are client-rendered and could not be verified; no retention duration published.
- Retention Exceptions
- unknown for the API. Separately, Meta trains on EU users' PUBLIC Facebook/Instagram posts + Meta AI interactions on a legitimate-interest (opt-out) basis — contested by regulators.
- Subprocessors
- Meta / partner inference infra (e.g. Cerebras, Groq — US)
- Transfer Mechanism
- none-stated
- Eu Procurement Ready
- fail
- Eu Procurement Reason
- No EU data residency for the hosted API; API-level DPA, retention, ZDR and certifications are unverifiable (legal pages are SPAs). Meta was the only major lab to REFUSE the EU AI Act GPAI Code of Practice (2025-07-18), and its EU consumer-post training resumed under a contested legitimate-interest basis. The durable sovereign path is self-hosting the open weights.
- Last Policy Change
- 2025-07-18T00:00:00.000Z
- Change Summary
- Meta declined to sign the EU AI Act GPAI Code of Practice, citing legal overreach — the only major lab to refuse.
- Prior Value
- No public stance on the Code; EU training paused 2024 after Irish DPC + noyb complaints, resumed Apr 2025 with an opt-out form.
- Training Evidence Url
- https://ai.meta.com/blog/llamacon-llama-news/
- Training Claim Basis
- stated
- Training Confidence
- high
- Residency Evidence Url
- https://techcrunch.com/2025/04/29/meta-previews-an-api-for-its-llama-ai-models/
- Residency Claim Basis
- inferred
- Residency Confidence
- medium
- Retention Evidence Url
- https://ai.developer.meta.com/legal/terms-of-service
- Retention Claim Basis
- inferred
- Retention Confidence
- low
- Subprocessors Evidence Url
- https://techcrunch.com/2025/04/29/meta-previews-an-api-for-its-llama-ai-models/
- Subprocessors Claim Basis
- inferred
- Subprocessors Confidence
- low
