Reading API service terms
A **literacy** method for reading vendor **service terms**, privacy centers, and product-term updates that govern hosted AI APIs. This is **not legal advice**. Primary docs change—**always re-read the current URL and note the date**. <!-...
What it is
A literacy method for reading vendor service terms, privacy centers, and product-term updates that govern hosted AI APIs. This is not legal advice. Primary docs change—always re-read the current URL and note the date.
<!-- IMAGE: checklist: parties → data use → AUP → region → indemnity carve-outs -->
Visual Spec & Architecture Diagram
API terms checklist: data use/training, retention, subprocessors, uptime, rate limits, indemnity, export, termination. Checkbox UI. Banner 'not legal advice—read live terms'.
Why it matters
Weights licenses do not replace API contracts. OpenAI Service Terms, Gemini API Additional Terms, AWS Service Terms (including AI/ML and Bedrock subsections), Microsoft product-term change pages, and Anthropic’s Privacy Center are where retention, prohibited uses, and commercial conditions live for hosted stacks.
How it works (plain)
When you open a terms page, hunt for:
- What counts as paid vs unpaid (data-use often differs).
- How prompts/outputs may be used (improve services? human review?).
- Retention / abuse monitoring windows.
- Geographic / age / medical restrictions.
- Competing model / reverse engineering limits.
- Indemnity scope and carve-outs (if any).
- Third-party models on a host (Bedrock: provider terms may also apply).
Literacy examples from fetches around 2026-08-14 (verify again before reliance):
- Gemini Additional Terms (effective March 23, 2026 per page): 18+; unpaid vs paid data-use differences; no competing-model development / weight extraction; medical-device style restrictions; region rules for EEA/CH/UK user-facing clients.
- AWS Service Terms §50 / Bedrock: AI Services definitions; abuse detection may store inputs/outputs for ~30 days for certain models; third-party model terms; customized model access limits.
- OpenAI Service Terms: govern Services alongside broader agreements; API use must follow docs; output IP indemnity with listed carve-outs (see live page).
- Anthropic Privacy Center: separates commercial vs consumer collections—start at the correct hub.
- Microsoft EAEAS product-term changes: track generative AI / Azure AI related licensing updates on the official change list.
Everyday example
Apartment lease vs house deed: both let you “use a kitchen,” but the rules about guests, sublets, and repairs differ. APIs are leases.
Try it
Pick one vendor. Paste five headings from their current terms into a note with today’s date. Mark which heading affects your data retention story.
Myths
- ⚠️ Myth: Marketing “we don’t train on your data” replaces reading terms.
- ✓ Reality: Tier, region, and product line matter; unpaid tiers often differ.
- ⚠️ Myth: One read at contract signing lasts forever.
- ✓ Reality: Terms and product sheets update—calendar a review.
Sources
- https://openai.com/policies/service-terms/ ↗
- https://ai.google.dev/gemini-api/terms ↗
- https://privacy.anthropic.com/en/ ↗
- https://aws.amazon.com/service-terms/ ↗
- https://www.microsoft.com/licensing/terms/en-US/product/changes/EAEAS ↗
- OpenAI data controls (companion literacy): https://developers.openai.com/api/docs/guides/your-data ↗
