AI Vendor Directory · Foundation model APIs

Will Google Gemini API sign a BAA?

Per Google Gemini API's published pages as of 2026-07-29: HIPAA BAA available (covered Google Cloud service). Check which plan or configuration the BAA covers before any PHI is sent.

What is Google Gemini API (Vertex AI)?

Gemini models served through Google Cloud's Vertex AI platform (and the separate Gemini Developer API). SMBs use it for chat, RAG over their data, and multimodal features under GCP terms.

Key facts: BAA, retention and training

  • Vendor: Google — https://cloud.google.com/vertex-ai
  • Signs a HIPAA BAA: BAA available (per vendor) (vendor wording: HIPAA BAA available (covered Google Cloud service))
  • Source for the BAA answer: https://cloud.google.com/security/compliance
  • Data retention: Vertex AI (Google Cloud terms): Google will not use customer data to train or fine-tune any AI/ML models without prior permission or instruction; Gemini inputs are cached up to 24h by default (disable caching for zero data retention) and prompts may be logged for abuse monitoring, with opt-out available.
  • Trains on customer data by default: no (Vertex AI; also Gemini Developer API PAID tier). UNPAID Gemini Developer API tier: yes
  • Source for the training answer: https://cloud.google.com/vertex-ai/generative-ai/docs/data-governance
  • Data residency: US; EU; Global (customer-selected Google Cloud regions; ML processing commitments per region)
  • Last verified: 2026-07-29

Sending PHI to Google Gemini API? What a hospital reviewer will ask

  • If your product sends protected health information to Google Gemini API, Google Gemini API is your subcontractor under HIPAA, so you need a signed BAA with it before any PHI flows, on the plan or configuration the BAA covers.
  • Reviewers ask where the data is stored, how long the vendor keeps it, and whether it trains on it. Keep the vendor’s answers (and the signed BAA) with your vendor records.

Does Google Gemini API train on your data?

Depends — varies by plan or setting. Vendor-stated default: no (Vertex AI; also Gemini Developer API PAID tier). UNPAID Gemini Developer API tier: yes. Vertex AI (Google Cloud terms): Google will not use customer data to train or fine-tune any AI/ML models without prior permission or instruction; Gemini inputs are cached up to 24h by default (disable caching for zero data retention) and prompts may be logged for abuse monitoring, with opt-out available. The separate Gemini Developer API (ai.google.dev) splits by tier: Paid Services — 'Google doesn't use your prompts... or responses to improve our products'; Unpaid Services — 'Google uses the content you submit... and any generated responses to provide, improve, and develop Google products and services,' including human review; Google warns not to submit confidential data on the unpaid tier. Note: the Vertex data-governance page did not render for automated fetch; wording verified live via search index of the vendor page on 2026-07-29.

What certifications does Google Gemini API hold?

  • SOC 1/2/3 (Google Cloud — Vertex AI in scope)
  • ISO/IEC 27001, 27017, 27018 (Google Cloud)
  • HIPAA BAA available (covered Google Cloud service)

Security, DPA, and subprocessor links

  • Security / trust page: https://cloud.google.com/security/compliance
  • Data processing agreement (DPA): https://cloud.google.com/terms/data-processing-addendum
  • Subprocessor list: https://cloud.google.com/terms/subprocessors

Optional, if you sell into the EU: your EU AI Act role when you build on Google Gemini API

Building on Vertex AI/Gemini typically makes you the provider of your downstream AI system under the EU AI Act, with Article 50 transparency duties for your end users; Google holds GPAI model-provider obligations for Gemini.

What to record in your vendor or AI system inventory

  • Provider: Google
  • Model type: third party api
  • Data typically flowing to the vendor: prompts, completions, grounding/RAG corpus data, images/video/audio inputs

Sources

  • https://ai.google.dev/gemini-api/terms
  • https://cloud.google.com/vertex-ai/generative-ai/docs/data-governance
  • https://cloud.google.com/terms/data-processing-addendum
  • https://cloud.google.com/terms/subprocessors

Disclaimer

Curated from vendor-published pages on the date above. Vendor terms change - verify against the cited sources before relying on a profile for a procurement decision. Last verified 2026-07-29. This directory is not legal advice and does not confirm any vendor’s HIPAA compliance; a BAA is between you and the vendor.

Frequently asked questions

Will Google Gemini API sign a BAA?

Per Google Gemini API's published pages as of 2026-07-29: HIPAA BAA available (covered Google Cloud service). Check which plan or configuration the BAA covers before any PHI is sent.

Does Google Gemini API train AI models on your data?

Depends — varies by plan or setting. Vendor-stated default: no (Vertex AI; also Gemini Developer API PAID tier). UNPAID Gemini Developer API tier: yes. Vertex AI (Google Cloud terms): Google will not use customer data to train or fine-tune any AI/ML models without prior permission or instruction; Gemini inputs are cached up to 24h by default (disable caching for zero data retention) and prompts may be logged for abuse monitoring, with opt-out available. The separate Gemini Developer API (ai.google.dev) splits by tier: Paid Services — 'Google doesn't use your prompts... or responses to improve our products'; Unpaid Services — 'Google uses the content you submit... and any generated responses to provide, improve, and develop Google products and services,' including human review; Google warns not to submit confidential data on the unpaid tier. Note: the Vertex data-governance page did not render for automated fetch; wording verified live via search index of the vendor page on 2026-07-29.

How long does Google Gemini API keep your data?

Per the vendor's published terms as of 2026-07-29: Vertex AI (Google Cloud terms): Google will not use customer data to train or fine-tune any AI/ML models without prior permission or instruction; Gemini inputs are cached up to 24h by default (disable caching for zero data retention) and prompts may be logged for abuse monitoring, with opt-out available.

What certifications does Google Gemini API hold?

Per the vendor's published pages as of 2026-07-29: SOC 1/2/3 (Google Cloud — Vertex AI in scope); ISO/IEC 27001, 27017, 27018 (Google Cloud); HIPAA BAA available (covered Google Cloud service).

Where does Google Gemini API store your data?

Vendor-listed data residency options as of 2026-07-29: US; EU; Global (customer-selected Google Cloud regions; ML processing commitments per region).

Selling into the EU: what is your EU AI Act role when you build on Google Gemini API?

Building on Vertex AI/Gemini typically makes you the provider of your downstream AI system under the EU AI Act, with Article 50 transparency duties for your end users; Google holds GPAI model-provider obligations for Gemini.