Express disclaimer
Terms expressly deny agency, partnership, fiduciary status, or name parties as independent contractors. Current examples are business/API/enterprise terms, not the surveyed consumer model terms.
Appendix G snapshot
Frontier-Model Terms Landscape
Appendix G reviews consumer-facing terms for Claude, ChatGPT, Gemini, and Grok as retrieved on April 19, 2026. The summary below is a dated posture map, not legal advice.
Four postures
Implicit risk allocation
Terms shift output reliance, professional-advice, warranty, confidentiality, liability, and user-action risk to the user without expressly accepting loyalty duties.
Silence / ambiguity
Terms materially lack both express disclaimer and strong risk-allocation clauses. No strong example appeared among the four surveyed consumer products.
Express acceptance
Terms voluntarily accept defined loyalty, care, obedience, disclosure, confidentiality, conflict-management, or confirmation duties. This is the design target proposed by the report.
Consumer-provider summary
All four surveyed consumer products were classified as implicit risk allocation. Paid consumer tiers did not materially change that posture.
| Provider / product | Effective or updated date | Classification | Snapshot finding |
|---|---|---|---|
| Anthropic Claude.ai / Claude Pro | October 8, 2025 | Implicit risk allocation | No express consumer fiduciary or no-agency clause found; user responsible for inputs/actions; independent confirmation and as-is terms. |
| OpenAI ChatGPT / ChatGPT Plus | January 1, 2026 | Implicit risk allocation | Consumer terms allocate output-use and professional-advice risk. Business Services Agreement separately contains an express no-agency clause. |
| Google Gemini / Gemini Advanced | May 22, 2024 | Implicit risk allocation | General Google terms and Gemini notices use as-is, professional-advice, liability-cap, and confidentiality-warning language. |
| xAI Grok / SuperGrok | April 10, 2026 | Implicit risk allocation | Consumer terms rely on sole-risk, user-responsibility, as-is, liability-cap, and high-stakes automation restrictions. |
Layer distinction
Frontier-model consumer terms are not the same as downstream AI-agent provider commitments. A downstream agent provider can add its own customer agreement accepting bounded duties, even when the upstream model provider does not.
Full ToS and Contract Snapshots
Dated local copies of every Terms-of-Service, Services Agreement, Acceptable Use Policy, Usage Policy, and Privacy Policy referenced in Appendix G of the report. Download date 2026-04-21 (Pacific). These are research-supporting snapshots and are not legal advice.
Source method. Anthropic and Google documents were fetched directly from their canonical URLs (HTTP 200). OpenAI and xAI documents were retrieved via the Wayback Machine because their canonical hosts returned HTTP 403 Cloudflare bot-protection challenges — the same limitation Appendix G §G.8 already records. Readers independently verifying the quoted clauses should use a standard browser from a residential IP for OpenAI and xAI.
Full index with per-document methodology, effective dates, and snapshot timestamps is in docs/terms/README.md.
Anthropic
| Document | Effective date | Download | Description |
|---|---|---|---|
| Consumer Terms of Service | Oct 8, 2025 | Direct fetch, 2026-04-21 | Governs both Claude.ai (free) and Claude Pro; Pro subscription provisions are §6 of the same document. Classification (ii) — implicit risk allocation. No express no-agency / no-fiduciary clause. Contains §4 reliance and responsibility language, §11 AS-IS warranty disclaimer, and §11 liability cap. |
| Usage Policy | — | Direct fetch, 2026-04-21 | Formerly the Acceptable Use Policy. Defines permitted and prohibited categories of use for Anthropic consumer and API services. |
| Privacy Policy | — | Direct fetch, 2026-04-21 | Governs collection, use, and retention of user data across Anthropic consumer services. |
OpenAI
| Document | Effective date | Download | Description |
|---|---|---|---|
| Terms of Use — United States (consumer) | Jan 1, 2026 | Wayback 2026-04-04, fetched 2026-04-21 | Governs ChatGPT (free) and ChatGPT Plus for US users. Classification (ii) — implicit risk allocation. Appendix G records a correction: the "no partnership, joint venture or agency" clause is in the business Services Agreement, NOT this consumer document. |
| Terms of Use — Rest of World | — | Wayback 2026-04-11, fetched 2026-04-21 | Non-US, non-EU jurisdiction variant of the consumer Terms of Use. |
| Europe Terms of Use | — | Wayback 2026-04-07, fetched 2026-04-21 | EU-jurisdiction variant of the consumer Terms of Use. |
| Usage Policies | — | Wayback 2026-04-12, fetched 2026-04-21 | Cross-product acceptable-use rules for all OpenAI services. |
| Policies Hub (index) | — | Wayback 2026-04-12, fetched 2026-04-21 | Index page linking to OpenAI's full set of legal and policy documents. Included for provenance. |
| Services Agreement (business / API / Enterprise) | Jan 1, 2026 | Wayback ~2026-04-21, fetched 2026-04-21 | Governs API, ChatGPT Enterprise, ChatGPT Business, developer/business services. Expressly does NOT apply to consumer ChatGPT / Plus. Classification (i) — express disclaimer. §16.8: "OpenAI and Customer are not legal partners or agents but are independent contractors." |
| Service Terms (business) | — | Wayback ~2026-04-21, fetched 2026-04-21 | Product-specific terms layered on the Services Agreement for business customers. |
| Document | Effective date | Download | Description |
|---|---|---|---|
| Terms of Service (general) | May 22, 2024 | Direct fetch, 2026-04-21 | Governs consumer Gemini (free) and Gemini Advanced. Classification (ii) — implicit risk allocation. AS-IS warranty disclaimer; professional-advice disclaimer (medical/legal/financial); user-account responsibility; liability cap (greater of $200 or 12 months of fees). |
| Gemini Apps Privacy Hub | — | Direct fetch, 2026-04-21 | Consumer-facing data-handling and transparency page. Contains the inaccuracy notice and the confidentiality warning quoted in Appendix G. |
| Google One Additional Terms | — | Direct fetch, 2026-04-21 | Governs paid Gemini Advanced (Google One AI Premium). Layered on the general Google Terms; no substantive duty-allocation delta vs. free tier. |
| Generative AI Prohibited Use Policy | — | Direct fetch, 2026-04-21 | Defines prohibited categories of use for Google's generative-AI products. |
xAI
| Document | Effective date | Download | Description |
|---|---|---|---|
| Consumer Terms of Service | Apr 10, 2026 | Wayback ~2026-04-21, fetched 2026-04-21 | Governs both free Grok and SuperGrok (~$30/mo, a correction over earlier $20/mo figure). Classification (ii) — implicit risk allocation. No express consumer no-agency clause. Prohibits "taking unauthorized actions on behalf of others" and "high-stakes automated decisions" affecting safety / legal / material rights / well-being. Liability cap (greater of fees paid or $100). |
| Enterprise Terms of Service | — | Wayback ~2026-04-21, fetched 2026-04-21 | Governs enterprise xAI customers. Does NOT govern consumer Grok / SuperGrok. Classification (i) — express disclaimer. Contains: "The parties to this Agreement are independent contractors. There is no relationship of partnership, joint venture, employment, franchise or agency created hereby between the parties." |
| Acceptable Use Policy | — | Wayback ~2026-04-21, fetched 2026-04-21 | Prohibited-use categories for xAI products. |
| Privacy Policy | — | Wayback ~2026-04-21, fetched 2026-04-21 | Governs data handling across xAI products. |
xAI — archived prior versions of Consumer Terms of Service
Appendix G §G.5 cites four dated prior versions for historical comparison. Each is a Wayback Machine snapshot captured on or near the publication date of that prior version.
| Prior version | Download | Description |
|---|---|---|
| Consumer Terms — prior version dated January 2, 2025 | Wayback 2025-01-02, fetched 2026-04-21 | Historical snapshot for ToS-drift tracking. |
| Consumer Terms — prior version dated February 14, 2025 | Wayback 2025-02-14, fetched 2026-04-21 | Historical snapshot for ToS-drift tracking. |
| Consumer Terms — prior version dated April 7, 2025 | Wayback 2025-04-07, fetched 2026-04-21 | Historical snapshot for ToS-drift tracking. |
| Consumer Terms — prior version dated June 9, 2025 | Wayback 2025-06-09, fetched 2026-04-21 | Historical snapshot for ToS-drift tracking. |
Canonical URLs, per-document snapshot timestamps, and sanity-check notes are in docs/terms/README.md.