Commitmentschangelog →

What AI vendors commit to, in their own words.

Seven commercial commitments tracked across AI vendors: acceptable use policy, no-train default, output ownership, IP indemnity program, retention default, deprecation notice, and safety framework version. Every cell is anchored to a primary source URL. No interpretation. The product is structured citation.

15
Vendors
7
Commitments per vendor
5 / 105
Verified cells
3
Comparing
Side-by-side compare

Comparing Anthropic, Google Vertex AI / Gemini, OpenAI. URL is shareable.

View full matrix →
Anthropic
Claude
AUP

Usage Policy (effective September 15, 2025)

verifiedsource ↗2025-09-15
No-train default

Anthropic may not train models on Customer Content (Inputs + Outputs) under Commercial Terms

verifiedsource ↗2025-06-17
Output ownership

Customer retains all rights to Inputs; owns Outputs

verifiedsource ↗2025-06-17
IP indemnity program

Pending verification (Claude Enterprise terms)

draftsource ↗
Retention default

API: 30 days for safety. Enterprise: 0 days available.

draftsource ↗
Deprecation notice

At least 60 days notice before retirement for publicly released models

verifiedsource ↗
Safety framework

Responsible Scaling Policy v3.3 (effective May 26, 2026)

verifiedsource ↗2026-05-26
Google Vertex AI / Gemini
Gemini, Vertex AI, Google DeepMind
AUP

Generative AI Prohibited Use Policy

draftsource ↗
No-train default

Vertex AI: customer data not used to train base models. Consumer Gemini: distinct policy.

draftsource ↗
Output ownership

Customer owns Outputs

draftsource ↗
IP indemnity program

Indemnified Generative AI Services (capped, named services)

draftsource ↗
Retention default

Vertex 30 days default; configurable lower

draftsource ↗
Deprecation notice

Ad-hoc per product. No standardized cross-product commitment.

draftsource ↗
Safety framework

Frontier Safety Framework (DeepMind)

draftsource ↗
OpenAI
ChatGPT, GPT-4o, o-series
AUP

Usage policies (current)

draftsource ↗
No-train default

API and Enterprise: not used for training by default. Consumer ChatGPT: opt-out available.

draftsource ↗
Output ownership

Customer assigned all right, title, and interest in Output

draftsource ↗
IP indemnity program

Copyright Shield (ChatGPT Enterprise, API)

draftsource ↗
Retention default

API: 30 days default. Zero-data-retention available for Enterprise.

draftsource ↗
Deprecation notice

Stated commitments: 6 months legacy / 3 months next-version / shorter for fine-tune deprecations

draftsource ↗
Safety framework

Preparedness Framework (v2, April 2025)

draftsource ↗

Methodology. Each cell quotes from a vendor primary source (Acceptable Use Policy, Commercial Terms, Trust Portal, Safety Framework, or equivalent). The product never paraphrases or interprets vendor language. It quotes and links.

How verification works. Cells marked pending are scaffolded with a best-effort summary and the source URL, but the exact clause has not yet been pasted into the quote field. The verification flow: open the source, find the relevant clause, paste the exact text into the cell, then flip the verified flag. Until a cell is verified it should not be cited in writing.

Found a cell that is wrong, stale, or missing? Email hello@anteroom.so.