| Trains on your data | No Checked 19 Oct 2026Source | No, unless you permit it Checked 21 Oct 2026Source | No, unless you permit it Checked 19 Oct 2026Source |
|---|
| Default retention | Under the default mode, retention follows the model. AWS may retain data for safety and abuse prevention. Setting store to false does not by itself guarantee zero retention. Checked 15 Nov 2026Source | For models sold by Azure, including Azure OpenAI, prompts and completions are not stored in the model. Abuse monitoring may store a sample for human review unless modified abuse monitoring is approved. Checked 15 Nov 2026Source | Implicit caching stores customer data and deletes it no later than 24 hours after it is written. Abuse-monitoring logs and grounding features have separate retention. Checked 15 Nov 2026Source |
|---|
| Zero data retention | Mode none is zero data retention: no request or response data is written to durable storage by AWS or shared with the model provider. Some models require retention, and zero data retention for those models is evaluated per account and per model. Checked 15 Nov 2026Source | Not stated by vendor Checked 21 Oct 2026Source | Requires customer configuration Checked 21 Oct 2026Source |
|---|
| SOC 2 | SOC 1, 2, and 3. Marketplace excluded. Checked 19 Oct 2026Source | Not stated by vendor Checked 21 Oct 2026Source | SOC 2 Type II. Includes Gemini Enterprise Agent Platform (formerly Vertex AI). Checked 19 Oct 2026Source |
|---|
| HIPAA | HIPAA-eligible services list. BAA required before PHI. Checked 19 Oct 2026Source | Not stated by vendor Checked 21 Oct 2026Source | BAA required. Gemini Enterprise Agent Platform is on the covered list. Checked 19 Oct 2026Source |
|---|
| FedRAMP | FedRAMP Class C (formerly Moderate) in East/West and Class D (formerly High) in GovCloud Checked 21 Oct 2026Source | Azure OpenAI is listed for FedRAMP High on Azure commercial and on Azure Government. The same page lists named Foundry tools. It does not mark every Foundry model. Checked 15 Nov 2026Source | FedRAMP High Checked 21 Oct 2026Source |
|---|
| EU residency | Geographic cross-Region inference keeps data processing within geographic boundaries such as the EU Checked 21 Oct 2026Source | Data Zone deployment processes data within the Azure EU Data Boundary. Not all models support all deployment types Checked 21 Oct 2026Source | The EU multi-region endpoint keeps processing inside EU member states. The United Kingdom and Switzerland are outside that endpoint. A global endpoint does not provide data residency guarantees. Checked 15 Nov 2026Source |
|---|
| Uptime SLA | 99.9% monthly uptime percentage. Service credits apply below 99.9%, and below 99.0% and 95.0%. The SLA page was last updated 4 October 2023. Checked 19 Oct 2026Source | 99.9% uptime for Foundry Models sold by Azure, including Azure OpenAI. Checked 21 Oct 2026Source | Gemini Online Inference API on Gemini Enterprise Agent Platform: 99.5% monthly uptime for generateContent and streamGenerateContent. Models designated for shorter availability are 95%. Checked 15 Nov 2026Source |
|---|
| Sub-processor notice period | 30 days Checked 21 Oct 2026Source | six months Checked 26 Oct 2026Source | 30 days Checked 21 Oct 2026Source |
|---|
| Objection window | Not stated by vendor Checked 21 Oct 2026Source | Not stated by vendor Checked 26 Oct 2026Source | 90 days Checked 21 Oct 2026Source |
|---|
| Last terms change | Last updated 4 October 2023 Checked 15 Nov 2026Source | SLA updated 1 October 2026 Checked 15 Nov 2026Source | SLA last modified 30 June 2026 Checked 15 Nov 2026Source |
|---|