1. What the document is: DeepSeek's consumer-facing Terms of Service, covering account terms, Input/Output ownership, content restrictions, and IP.
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 – Terms of Service
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/Terms of Service/ (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,feedback). 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’s consumer-facing Terms of Service, covering account terms, Input/Output ownership, content restrictions, and IP.
2. Input rights: Mirrors the Developer Terms exactly: “You retain any rights, title, and interests… in the Inputs you submit” and “We assign any rights, title, and interests… in the Outputs of the Services to you” (§4.2), with the same explicit permission to use Inputs/Outputs for “training other models (such as model distillation)” — confirming this permissive, non-restrictive training stance applies to consumer users too, not just API developers. This makes it the dataset’s clearest example of a provider explicitly authorizing competing-model training from its own Output, the opposite of OpenAI’s and Anthropic’s bans.
3. Output restrictions: §3.1 requires that anyone publishing/disseminating Service-generated Output must “clearly indicate that the output content is generated by artificial intelligence” (ai disclosure) and verify factual accuracy before sharing. §1.6 (added 2026-03-27) discloses a public Dialogue-sharing feature (via shareable URL) and that DeepSeek applies “industry-standard anti-crawling mechanisms (the robots exclusion protocol)” to shared content, while acknowledging scraping risk “cannot be entirely eliminated.”
4. Non-explicit predatory clauses: Feedback: “We appreciate your feedback, and you agree that we may use it without restriction or compensation to you” — same broad, uncompensated pattern as every other provider in this dataset. §3.3 separately grants DeepSeek a broad content-review right (“technical means to review the behavior… of users,” “risk filtering mechanisms,” “databases for illegal content features”) for compliance purposes — narrowed slightly in the 2026-03-27 revision (dropped explicit mention of “reviewing inputs and outputs,” though the broader review right remains).
5. Regulatory references: None found in this document specifically (the Privacy Policy carries the GDPR Article 27 representation badges) — only generic “Export Control and Sanctions Laws” and “applicable laws and regulations” references.
6. Regional variation: Not explicitly addressed in this Terms of Service document (no EEA/UK-specific carve-out), though the separate Privacy Policy discloses China-based data storage for all users including EU users.
7. Key risk to users: The IP clause (§6.1) reserves all rights in “model weights and parameters” etc. to DeepSeek while explicitly excluding Inputs/Outputs from that reservation — consistent with the user-favorable ownership stance. The most consequential clause for ordinary users is the public-sharing/web-crawler exposure risk (§1.6), since DeepSeek explicitly tells users this risk “cannot be entirely eliminated” despite its anti-crawling mitigations.
Flagged Keywords & Risks (LLM-assigned)
output ownership,input license— the same Input/Output ownership-and-use clause as the Developer Terms, explicitly permitting “training other models (such as model distillation).” Why it matters: confirms DeepSeek’s permissive training stance covers consumer users too, not just API developers — a genuine outlier next to OpenAI’s and Anthropic’s competing-model bans.ai disclosure— users must disclose that Output is AI-generated when publishing or sharing it. Why it matters: consistent with the AI-disclosure pattern found across this dataset’s other providers.feedback— DeepSeek can use any feedback a user gives however it wants, with no restriction and no payment. Why it matters: identical pattern to every other provider’s Feedback clause in this dataset.
Regulatory & Research Context
Pandit et al. (2026), who directly coded DeepSeek’s terms in their FAccT study, found that “all terms in our analysis contained language that explicitly discards assurances regarding the quality, availability and appropriateness of the service” — consistent with this page’s §3.1 disclaimer requiring users to “proactively verify the authenticity and accuracy” of Output before sharing it. Davidson et al. (2026), who also studied DeepSeek directly, note that DeepSeek’s disputes are located in Hangzhou and that some of its terms “override their legal jurisdictions regardless of user location,” a jurisdictional posture that compounds the §3.3 content-review right this page flags as a feedback-adjacent predatory clause.
Changes Summary
| Date | What changed |
|---|---|
| 2025-11-11 | Baseline version (“Last Update: April 28, 2025”). |
| 2026-03-27 | Genuine addition: new §1.6 disclosing the public Dialogue-sharing-via-URL feature and DeepSeek’s robots-exclusion-protocol anti-crawling mitigation. Several legal-standard tightenings (e.g., “intentional or negligent” → “intentional or gross negligent” conduct exception; “we will provide the fullest cooperation” → “commercially reasonable efforts”; account-suspension explanation/objection standard changed to “at our sole discretion”); §3.3’s content-review clause narrowed slightly (dropped explicit “reviewing inputs and outputs” wording, though the broader review right remains); heading-style formatting changes throughout. No substantive change to the Input/Output ownership, training-permission, AI-disclosure, or Feedback clauses. |
Version History
2025-11-11
- Explicit AI clause: YES
- Non-explicit predatory: YES
- Flagged keywords:
output ownership,input license,ai disclosure,feedback
Clause: output ownership, input license
4.2 Subject 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. Due to the nature of our Services and artificial intelligence generally, Outputs may not be unique and other users may receive similar Outputs from our Services. Our assignment above does not extend to other users’ Outputs. (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., as long as such usage is legal and adhere to these Terms.
Clause: ai disclosure
3.1 […] If you publish or disseminate outputs generated by the Services, you must: (1) proactively verify the authenticity and accuracy of the output content to avoid spreading false information; (2) clearly indicate that the output content is generated by artificial intelligence, to alert the public to the synthetic nature of the content; (3) avoid publishing and disseminating any output content that violates the usage specifications of these Terms.
Clause: feedback
We appreciate your feedback, and you agree that we may use it without restriction or compensation to you.
2026-03-27
- Explicit AI clause: YES
- Non-explicit predatory: YES
- Flagged keywords:
output ownership,input license,ai disclosure,feedback
All flagged clauses unchanged from 2025-11-11 (see above) — this date’s genuine addition was the public-sharing/anti-crawling disclosure (§1.6, not separately flagged as it’s a security/risk disclosure rather than a rights clause), plus legal-standard wording tightenings elsewhere.
Clause: new disclosure (not separately flagged) — public Dialogue sharing
1.6 Users may also share the Inputs and Outputs information (together referred to as “Dialogues”) by generating a unique URL. Anyone who has access to a shared link can view the linked Dialogues. When using such feature, any Dialogues you share with others, if subsequently published on public networks, may be at risk of being obtained by third parties through technical means such as web crawlers. We have implemented industry-standard anti-crawling mechanisms (the robots exclusion protocol) to prevent third parties from scraping user-publicly shared content. However, due to the technical limitations commonly existing within the industry, such risks cannot be entirely eliminated.