Data processing agreement in MLOps and model serving
Without a published DPA you cannot review the contract before you talk to sales.
6 of 34 companies we measured here publish it, that is 18%. Last checked 2026-10-18.
18% here against 11% across every category we measure. MLOps and model serving sits above the overall share for this criterion.
This page counts companies, not product entries: a company with two products in this category is one company here and two rows below.
Back to Data processing agreement across all categories, or to MLOps and model serving.
- Found6
- Not found at checked addresses28
- Not measured3
37 companies in total. Unmeasured companies are excluded from the published share.
Published
Published and readable without a login.
| Vendor | What we found |
|---|---|
| Baseten | https://www.baseten.co/dpa/Checked |
| Dataiku | https://www.dataiku.com/legal/dataiku-cloud/data-processing-addendumChecked |
| H2O.ai | https://h2o.ai/legal/dpa/Checked |
| Modal | https://modal.com/legal/dpaChecked |
| RunPod | https://www.runpod.io/legal/data-processing-agreementChecked |
| Vast.ai | https://vast.ai/data-processing-agreementChecked |
No data processing agreement found.
We reached at least one address for each of these vendors and none of them carried it. That is what we found at the addresses we checked, on the date we checked them, and not a statement about what the vendor has.
| Vendor | What we found |
|---|---|
| Anyscale | 8 addresses checkedChecked |
| BentoML | 8 addresses checkedChecked |
| ClearML | 8 addresses checkedChecked |
| Comet | 8 addresses checkedChecked |
| CoreWeave | 8 addresses checkedChecked |
| DataChain | 8 addresses checkedChecked |
| DataRobot | 8 addresses checkedChecked |
| Evidently | 8 addresses checkedChecked |
| Fireworks AI | 8 addresses checkedChecked |
| Flyte | 8 addresses checkedChecked |
| Genesis Cloud | 8 addresses checkedChecked |
| Kedro | 8 addresses checkedChecked |
| Lambda | 8 addresses checkedChecked |
| Metaflow | 8 addresses checkedChecked |
| Mistral AI | 8 addresses checkedChecked |
| Nebius | 8 addresses checkedChecked |
| Northern Data | 8 addresses checkedChecked |
| Nscale | 8 addresses checkedChecked |
| Paperspace | 8 addresses checkedChecked |
| Replicate | 8 addresses checkedChecked |
| Seldon | 8 addresses checkedChecked |
| StackIt | 8 addresses checkedChecked |
| Superwise | 8 addresses checkedChecked |
| Together AI | 8 addresses checkedChecked |
| Union | 8 addresses checkedChecked |
| Valohai | 8 addresses checkedChecked |
| Wallaroo | 8 addresses checkedChecked |
| ZenML | 8 addresses checkedChecked |
Not measured
We could not look at these, or their robots.txt asks us not to. That says something about our measurement, not about them, and they are left out of the percentage above.
| Vendor | What we found |
|---|---|
| Domino | The vendor’s robots.txt excludes us from at least one of these addresses, so this was not measured.Checked |
| Lightning AI | The vendor’s server blocked or rate-limited our crawler, so this was not measured.Checked |
| Weights & Biases | This vendor was not measured because our run stopped waiting for it.Checked |
Other criteria in MLOps and model serving
- Subprocessor list
- Data location stated
- Status page with history
- Public pricing
- Certifications named
- Uptime SLA with a figure
- HTTPS enforced with HSTS
- No third-party tracking before consent
- security.txt (RFC 9116)
- Notice before subprocessors change
- Named privacy contact
Data processing agreement in related categories
Found something we missed? Tell us. A correction moves that vendor into our next nightly run. That run reads the same fixed addresses we check for every vendor, not the address you send.