← All pages

1. What the document is: A short "Llama API Data Commitments" disclosure panel, distinct from (and far less detailed than) the GDPR-style Data Processing Addendum/Privacy Policy documents seen from other providers in this dataset.

What this wiki found — complete, every page

Presence facts, not verdicts — each flagged term links to its definition and the exact clause on this page. Absence of a badge means the term isn't currently flagged here, not that the page is risk-free.

Independent reviews, where they had something to say

Llama API – Privacy Policy

Dataset: GenGA (Generative AI Governance Archive) — 11 AI services, 2025–present

GenGA (Generative AI Governance Archive) is this wiki’s Generative-AI dataset: raw policy snapshots captured by the Open Terms Archive genai-eu project across 11 GenAI providers (ChatGPT, Claude.ai, DeepSeek, Google Generative AI Services, Le Chat, Llama API, Meta AI, Microsoft Copilot, Perplexity, Qwen Chat, xAI). Unlike PGAv2, GenGA has no pre-tagged risk scores — all risk scoring and keyword tagging on these pages is LLM-assigned by direct reading, spanning 2025–present.

Note: This page contains documented policy clauses. Risk assessment is qualitative and context-dependent. For analysis of patterns across platforms, see:

• AI Act Substance Matrix
• Tier Bifurcation
• Scraper Enforcement Gap
• Methodology

Source: sources/GenGA/Llama API/Privacy Policy/ (raw Markdown captures, Open Terms Archive genai-eu dataset — no pre-tagged JSONL exists for this source) Last updated: 2026-04-15 (substantive content unchanged since baseline; see methodology note)

Note on methodology: No pre-computed risk_score/keyword fields exist for this source. Keyword tags and clause analysis on this page are LLM-assigned. Treat as first-pass analysis, not externally verified ground truth. This wiki’s risk-scoring system was retired project-wide on 2026-06-21 (see methodology.md §4) — no page, GenGA included, computes or displays a numeric risk score.

Document-bundling note: This capture bundles a short, Llama-API-specific “Data Commitments” panel with Meta’s entire general-purpose consumer Privacy Policy (Facebook/Instagram/Messenger) appended below it. Only the Llama-specific panel is analyzed here — the bundled general Meta policy is out of scope (covered by Meta’s own separate, non-Llama-specific disclosures).

Content-hash dedup: 108 raw captures collapse to 77 distinct hashes, but only 5 cosmetic variants of one unchanging Llama-specific panel (breadcrumb-link and proxy-URL changes only — never a substantive wording change) plus ~34 transient scrape-error captures (“page isn’t available”) interspersed throughout the date range. The remaining hash variation comes entirely from the bundled general Meta policy being updated on its own independent schedule, and from rotating Facebook CDN signed-URL parameters in the page footer — both out of scope per the bundling note above.

Overview

1. What the document is: A short “Llama API Data Commitments” disclosure panel, distinct from (and far less detailed than) the GDPR-style Data Processing Addendum/Privacy Policy documents seen from other providers in this dataset.

2. Input rights: Unconditional no-training commitment, stated as an absolute with no tier-based exception and no opt-in/opt-out toggle anywhere in any of the 108 captures: “Meta does not train its AI models on Llama API Customer Content or Customer Data.” This is the strongest, simplest no-training statement in the entire GenGA dataset — even stronger in framing than Anthropic’s or Llama API’s own Terms of Service equivalent, since no exception categories (Feedback, moderation-flagged content, Labs/experimental models, etc.) are listed at all.

3. Output restrictions: Not addressed in this panel.

4. Non-explicit predatory clauses: None found. A closed, four-item permitted-use list (provide the service; understand usage; enforce the AUP; comply with legal obligations) with an explicit no-advertising, no-sale, no-personalization commitment: “Customer Content and Customer Data are not used for personalization of consumer Meta Products or advertising… Meta will not sell any customer or user data.”

5. Regulatory references: None found — no GDPR article citations, no SCCs, no UK GDPR, no adequacy-decision language, and no EU AI Act reference anywhere in the Llama-specific panel (in sharp contrast to the GDPR/MGPT apparatus found in the companion Llama API Terms of Service document).

6. Regional variation: None found in this panel — notably, the EU-specific multimodal-model access restriction documented in Llama API’s Terms of Service has no counterpart here; this document treats all users identically regardless of region.

7. Key risk to users: Low for the stated commitments themselves, but the document’s thinness is itself the risk: no retention period is disclosed, no subprocessor list is provided, and no children’s/sensitive-data-specific clause exists for the Llama API — encryption (TLS 1.2/1.3, ChaCha20-Poly1305/AES-GCM at rest) and PCI-DSS payment-data handling are the only concrete technical commitments; everything else (e.g., “strict access controls,” “logically separated”) is generic, unquantified language.

Flagged Keywords & Risks (LLM-assigned)

• train AI/models (unconditional negative) — “Meta does not train its AI models on Llama API Customer Content or Customer Data.” Why it matters: the cleanest, most absolute no-training commitment in this dataset — useful as the strongest reference point in the cross-provider comparison table.

Regulatory & Research Context

Llama API was not among the providers directly studied by Davidson et al. (2026) or Pandit et al. (2026) in this batch, but Pandit et al.’s general finding — that GenAI terms typically leave users facing “lack of necessary information” — applies by analogy to this page’s own thinness, since the unconditional no-training statement comes with no disclosed retention period, no subprocessor list, and no children’s/sensitive-data clause. Edwards et al. (2025) describe privacy disclosures that “requir[e] reading comprehension abilities at university level” as part of the broader “platformisation paradigm”; this panel’s brevity is the inverse problem — an absolute commitment stated with unusual clarity, but unaccompanied by the technical and procedural detail found in other providers’ DPAs in this dataset.

Changes Summary

Version History

2025-11-11

• Explicit AI clause: YES
• Non-explicit predatory: NO
• Flagged keywords: train AI/models (unconditional negative)

Clause: train AI/models

Meta does not train its AI models on Llama API Customer Content or Customer Data.

Clause: definitions (not separately concept-tagged)

Customer Data comprises your account data, roles, and profile fields. Customer Content comprises both the inputs you submit to the model (prompts) and the outputs you receive based on your inputs (model responses).

Clause: permitted uses (not separately concept-tagged)

Your Customer Content and Customer Data are used only: 1. To provide Llama API services. 2. To understand how our services are used. 3. To ensure adherence to the Acceptable Use Policy. 4. To comply with legal obligations.

Clause: no-sale/no-advertising commitment (not separately concept-tagged)

Your Customer Content and your Customer Data are not used for any purposes other than those described in our permitted purposes. Customer Content and Customer Data are not used for personalization of consumer Meta Products or advertising, and personal data collected from the use of Llama API products is not used to personalize ads. […] Meta will not sell any customer or user data as defined in the Terms of Service.

Clause: encryption (not separately concept-tagged)

All data transmitted between Llama API and backend servers is encrypted with the industry-standard TLS 1.2 and TLS 1.3 protocols. […] this data is encrypted at rest using strong symmetric encryption algorithms such as ChaCha20-Poly1305 (XChaPoly) and AES-GCM.

This baseline text is unchanged, in substance, through the final 2026-04-15 capture — all 108 raw files in this dataset either reproduce it byte-for-byte, show only cosmetic navigation/link changes, or are transient scrape-error pages.

2026-04-15

All clauses unchanged from 2025-11-11 (see above).