1. What the document is: DeepSeek Open Platform's Terms of Service for API/developer access to DeepSeek Chat models.
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.
DeepSeek – Developer Terms
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-euproject 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:
Source: sources/GenGA/DeepSeek/Developer Terms/ (raw Markdown captures, Open Terms Archive genai-eu dataset — no pre-tagged JSONL exists for this source)
Last updated: 2026-03-27
Note on methodology: No pre-computed
risk_score/keywordfields exist for this source. Keyword tags and clause analysis on this page are LLM-assigned, using the same rubric/vocabulary as the PGAv2 pages plus new GenAI-specific tags (output ownership,input license,ai disclosure). Treat as first-pass analysis, not externally verified ground truth. This wiki’s risk-scoring system was retired project-wide on 2026-06-21 (seemethodology.md§4) — no page, GenGA included, computes or displays a numeric risk score.
Overview
1. What the document is: DeepSeek Open Platform’s Terms of Service for API/developer access to DeepSeek Chat models.
2. Input rights: “You retain any rights, title, and interests—if any—in the Inputs you submit” and “We assign any rights, title, and interests—if any—in the Outputs of the Services to you” — the same user-protective ownership pattern found across OpenAI’s and Anthropic’s commercial terms.
3. Output restrictions: Notably the opposite of every other provider in this dataset: §4.2(3) lets developers use Inputs and Outputs for “a wide range of use cases, including personal use, academic research, derivative product development, training other models (such as model distillation), etc.” Where OpenAI and Anthropic both ban using their Services’ Output to train competing models, DeepSeek’s Developer Terms allow it. §5.1 keeps DeepSeek’s IP ownership over the model itself (parameters, algorithms, code, framework) intact, so this permission covers Output content only, not the model weights or code.
4. Non-explicit predatory clauses: §8.1 requires developers to “clearly disclose to your end users that the Output content is generated by AI” — an ai disclosure mandate, consistent with the pattern found in OpenAI’s and Anthropic’s policies.
5. Regulatory references: None found — no GDPR, EU AI Act, or any other named regulation, despite this being the EU-region capture of the dataset. Only generic references to “applicable law,” “Export Control and Sanctions Laws,” and “competent authorities.”
6. Regional variation: Not explicitly addressed in this document (no EEA/UK-specific carve-out language, unlike OpenAI’s and Anthropic’s equivalent contracts) — notable given this is the genai-eu dataset.
7. Key risk to users: Low on the training-permission front — this is a developer-favorable permission, not a restriction. The real risk is elsewhere: a broad all-caps disclaimer that Output may not be “ACCURATE, UP-TO-DATE, RELIABLE, NON-INFRINGING OR SECURE” (§3), combined with a corporate-structure change (the 2026-03-27 update drops “Beijing DeepSeek Artificial Intelligence Co., Ltd.” as a joint operator, leaving only the Hangzhou entity) — worth tracking for what it could mean for liability down the line.
Flagged Keywords & Risks (LLM-assigned)
output ownership,input license— developers keep ownership of their Inputs, and DeepSeek assigns them ownership of the Outputs — but both are covered by a broad permitted-use list that explicitly includes training other models, including via distillation. Why it matters: this is the first provider in the dataset whose terms authorize training a competing model on its Output rather than ban it — the inverse of OpenAI’s and Anthropic’s approach, and a genuinely distinguishing data point in the cross-provider comparison.ai disclosure— developers must disclose Output is AI-generated to their end users. Why it matters: consistent with the transparency pattern found across this dataset’s other providers.
Regulatory & Research Context
Davidson et al. (2026), who directly studied DeepSeek’s Terms among their five-provider sample, frame their analysis around “regulatory gray areas” that create uncertainty for legitimate use; the §4.2(3) clause flagged above — explicitly permitting “training other models (such as model distillation)” — is the dataset’s clearest case of a provider’s terms authorizing exactly the kind of competing-model use other providers’ terms exist to forbid, making it a useful contrast case for that gray-area framework. Pandit et al. (2026), who also coded DeepSeek directly, note generally that GenAI terms create “significant imbalance of power” between provider and user; DeepSeek’s developer-favorable distillation permission is a rare instance in this dataset where the balance tips the other way.
Changes Summary
| Date | What changed |
|---|---|
| 2025-11-11 | Baseline version (“Last Update: January 20, 2025”). |
| 2026-03-27 | Corporate-structure narrowing: “jointly owned and operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., Beijing DeepSeek Artificial Intelligence Co., Ltd. and their affiliates” → “owned and operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd.” (Beijing entity dropped); minor grammar/formatting edits throughout (heading style, link formatting, capitalization of “Services”); §1.2 dropped a clause about assisting developers with “relevant obligations stipulated by laws and regulations.” No substantive change to the Output-ownership, training-permission, or AI-disclosure clauses. |
Version History
2025-11-11
- Explicit AI clause: YES
- Non-explicit predatory: NO
- Flagged keywords:
output ownership,input license,ai disclosure
Clause: output ownership, input license
4.2 Subject to your compliance to applicable law and our Terms, you have the following rights regarding the Inputs and Outputs of the Services: (1) You retain any rights, title, and interests—if any—in the Inputs you submit; (2) We assign any rights, title, and interests—if any—in the Outputs of the Services to you; (3) You may apply the Inputs and Outputs of the Services to a wide range of use cases, including personal use, academic research, derivative product development, training other models (such as model distillation), etc.
Clause: ai disclosure
8.1 Any and all Outputs provided by this Service are generated by the foundation model and shall not constitute DeepSeek’s legal declaration of intent; neither shall the Outputs represent DeepSeek’s views or opinions in any event. […] You shall clearly disclose to your end users that the Output content is generated by AI, and may contain errors or omissions and are for reference only.
2026-03-27
- Explicit AI clause: YES
- Non-explicit predatory: NO
- Flagged keywords:
output ownership,input license,ai disclosure
output ownership/input license/ai disclosure unchanged from 2025-11-11 — this date’s changes were the corporate-structure narrowing and minor grammar/formatting, not these flagged clauses.