The Short Answer
Gemini is not HIPAA compliant in its consumer form; it can be used in HIPAA-regulated work only under a signed BAA on Google Cloud, with the covered entity carrying the compliance obligation. The alternative removes the question: on ibl.ai you own all the code and the data, so PHI never leaves your perimeter — model-agnostic across any LLM, with no per-seat pricing.
The Direct Answer
Gemini — Google's frontier model family — is not HIPAA compliant in its consumer form. But Gemini can be used in HIPAA-regulated workflows through one specific route: Vertex AI on Google Cloud Platform, with a Business Associate Agreement (BAA) signed with Google, on a HIPAA-aligned Google Cloud account.
Consumer Gemini at gemini.google.com is not covered. Gemini in the consumer Google Workspace tiers is not covered. Gemini for Workspace Enterprise can be configured for HIPAA workloads under a BAA that names the covered services explicitly.
That is the legal layer. The deeper question — the one your CISO and your HIPAA compliance lead are going to ask after the BAA is signed — is where Protected Health Information actually flows during inference, who controls the audit trail, and what your continuity plan looks like if Google changes terms or deprecates a model.
What HIPAA Asks of an AI Inference Endpoint
HIPAA's Security Rule applies to every system that touches PHI. An AI inference endpoint that receives a prompt with patient data is a covered system. The Security Rule requires:
- Access control — only authorized users submit PHI prompts.
- Audit controls — every PHI prompt is logged and reviewable.
- Integrity — PHI is not tampered with in flight or at rest.
- Transmission security — PHI is encrypted hop-to-hop.
- Authentication — identities are verifiable across the chain.
A BAA is the legal acknowledgement that the business associate will meet these safeguards. The technical implementation determines whether the safeguards survive an audit.
What Google's BAA Covers, in Practice
Google Cloud's BAA covers a specific list of "Covered Services" that the customer can use to process PHI. The list is documented on Google's HIPAA compliance page and evolves as Google adds services. The relevant entries for AI buyers in 2026 are:
- Vertex AI — Google's managed AI platform, where Gemini models are exposed as API endpoints. Inference runs inside the customer's Google Cloud project, in the customer's chosen region, under the customer's IAM controls.
- Specific Workspace services for Gemini for Workspace Enterprise, named in the BAA.
- Cloud Healthcare API for FHIR/HL7 integration alongside Vertex inference.
What the BAA does not cover:
- Consumer Gemini at gemini.google.com.
- Gemini in any free or non-enterprise Workspace tier.
- Any usage that routes through unsupported APIs or geographies.
- What your downstream applications do with the response.
The Architectural Reality of Vertex AI for HIPAA
Vertex AI is the cleanest BAA-covered route for Gemini. Inference happens inside your Google Cloud project, in your VPC, in a region you choose. The data path is Google's — Google operates the infrastructure — but the project boundary, the IAM, the audit logging, and the regional residency are configured by you.
This is genuinely strong for HIPAA. For organizations already standardized on Google Cloud, it is often the right answer for routine clinical AI workloads.
The trade-offs are the same trade-offs every hyperscaler AI managed-service decision carries:
- You are coupled to one hyperscaler for that workload. Moving to Vertex AI is a decision that lives at the architecture-review level, not the prompt level.
- The audit logs flow into Google's logging stack first. You can stream them to your SIEM, but the canonical source is Google's Cloud Logging.
- Model lineup is Google's. When Google deprecates a Gemini variant or rebases a fine-tune, your clinical workflow has to follow.
- Costs scale with token usage. At hospital-system scale, the per-token Vertex pricing is not the same as the per-seat consumer pricing, and the budget conversation gets serious.
For many organizations this is acceptable. For the ones with the highest PHI sensitivity, an additional layer is worth considering.
The Sovereign-AI Posture
A self-hosted AI platform changes the question from "which hyperscaler's BAA covers our AI usage" to "PHI never leaves our perimeter unless we explicitly choose to send it."
The architecture is:
- The AI platform runs inside the covered entity. Not in Google's data center, not in a managed SaaS — in your VPC, your data center, or your air-gapped environment, under your network and IAM.
- The model is your choice, per workload. Gemini through Vertex AI's BAA for non-PHI workloads when frontier quality matters. Open-weights models (Llama, Mistral, Qwen, Phi) running on your GPUs for PHI-touching workloads.
- The governance layer is yours. Prompts, responses, model invocations all log into your SIEM. The BAA chain covers exactly what flowed externally, not your full audit surface.
This is the model the ibl.ai platform implements. The customer receives the complete source code on day one and runs the platform inside their infrastructure. Gemini, when used, is reached through Vertex AI under Google's BAA — but the ibl.ai platform sits in front of it, captures the audit trail in the customer's SIEM, and routes PHI-heavy workloads to a local open-weights model instead.
For the build-and-buy posture compliance teams want: a production-grade platform on day one (no engineering green-field), full source-code ownership (no vendor lock-in if priorities change), and routing flexibility per workload (frontier when appropriate, local when sensitive).
A Practical Decision Framework
For a hospital, health system, or payer evaluating Gemini for clinical use, the practical decision tree is:
- PHI-heavy workload? Self-hosted platform with a local open-weights model. No external inference path. Audit logs stay in your SIEM.
- Sometimes-PHI workload? Route through a self-hosted platform that can call Gemini via Vertex AI when needed, with the audit trail captured locally before any hyperscaler hop.
- Definitely non-PHI workload? Any BAA-covered Gemini route is fine — Vertex AI is the cleanest if you are already on Google Cloud.
The point of the sovereign posture is that you do not have to choose one model, one architecture, or one hyperscaler. You choose per workload, and the governance layer stays constant.
What to Take Away
- Gemini is not HIPAA compliant by default.
- Gemini can be used in HIPAA workflows through Vertex AI on Google Cloud under Google's BAA, and through specific Gemini-for-Workspace Enterprise services named in the BAA.
- Consumer Gemini and non-enterprise Workspace tiers are not covered.
- A BAA is the contractual layer; the technical layer — where PHI lives, who owns the audit trail, what your continuity plan is — is what HIPAA auditors actually test.
- The strongest HIPAA posture is a self-hosted AI platform with model choice per workload, local audit logs, and full source-code ownership.
See how ibl.ai's self-hosted and private LLM platform handles HIPAA-aligned Gemini routing, and how the Healthcare solution is structured around BAAs, data residency, and SIEM integration. The Google Gemini service page covers the Vertex AI deployment path specifically.
Evaluating other providers? The companion guides cover them: Is ChatGPT HIPAA compliant? and Is Claude HIPAA compliant?.