AI Vendor Data-Trust Index live
A transparent, source-cited comparison of how AI products handle your data — does the vendor train on it, is zero data retention available, can you opt out. Every value carries the exact policy quote, source URL, and date it was checked.
Every value is checked against the vendor's live policy and carries a verbatim quote, source URL, and date. Ambiguous cases are marked unclear rather than guessed. An automated verifier re-checks weekly; changes are logged. Spotted an error? Send a correction →
| Vendor | Trains on your data | Zero retention | Opt-out | Verdict |
|---|---|---|---|---|
| Amazon — Amazon Bedrock | No | Yes | N/A | Favorable |
| Anthropic — Claude API | No | Unclear | N/A | Mixed |
| GitHub — Copilot Business | No | Yes | N/A | Favorable |
| Google — Vertex AI (enterprise) | No | Yes | N/A | Favorable |
| Google — Gemini app (consumer) | Yes, unless you opt out | Unclear | Yes | Caution |
| Meta — Llama (open-weight) | Unclear | Unclear | Unclear | Unclear |
| Microsoft — 365 Copilot | No | Unclear | N/A | Mixed |
| Microsoft — Azure OpenAI Service | No | Yes | N/A | Favorable |
| Mistral AI — Mistral API | No | Yes | N/A | Favorable |
| OpenAI — API | No | Yes | N/A | Favorable |
| OpenAI — ChatGPT (consumer) | Yes, unless you opt out | Unclear | Yes | Caution |
What this answers
One plain question with a cited answer: if I send my data to this AI product, what happens to it? The headline fields are whether the vendor uses your inputs to train its models, whether a zero-data-retention option exists, and whether you can opt out. "Unclear" is used deliberately — a missing or ambiguous policy is never rounded up to a favorable answer.
How each value is established
Each field is extracted from the vendor's published policy and stored with the exact supporting quote, the source URL, a confidence level, and the date it was checked. A value changes only when the source text changes, and every change is dated. Nothing is inferred beyond what the policy states. An automated verifier re-checks the sources weekly: it only publishes a definite value when a verbatim quote from the live policy supports it, and holds the current human-verified value otherwise. Ambiguous or unconfirmable fields are marked unclear, never guessed.
How an AI agent can use it
Every response is source-cited with the quote and URL behind each value:
| Access | Endpoint | Cost |
|---|---|---|
| Per-vendor verdict + headline fields | GET trust.oscar-lopez.com/v1/ai-trust/headline?vendor=ID | free |
| Compare one field across vendors | GET trust.oscar-lopez.com/v1/ai-trust/compare?field=KEY | free |
| Full cited record (all fields) | GET trust.oscar-lopez.com/v1/ai-trust/vendor?vendor=ID | $0.02 (x402) |
| Bulk export (all vendors) | GET trust.oscar-lopez.com/v1/ai-trust/all | $0.02 (x402) |
| MCP tools | /mcp → headline & compare (free), full record (paid) | free / $0.02 |
Frequently asked
Is this legal advice?
No. It is a factual, source-cited summary of public vendor policies, not legal advice. Always confirm against the linked source and your own contract before relying on it.
How current is it?
Each value shows the date it was last checked. An automated verifier re-checks the live policies weekly, and every change is dated and logged.
Found something wrong?
Send the vendor, field, and the correct policy quote/URL to [email protected]. Corrections are dated and published.
Who stands behind it?
Oscar E. López, Ph.D. (Information Systems, UMass Amherst) — whose research includes online privacy and trust. It is an independent project, not affiliated with any listed vendor.