Blog

Is Google Gemini HIPAA compliant?

Is Google Gemini HIPAA compliant?

Google Gemini can be used in a HIPAA-compliant environment when the specific service is covered by Google’s Business Associate Agreement (BAA) and the healthcare organization implements the required safeguards. Google identifies certain Gemini services as eligible for coverage under its HIPAA BAA, including specified Google Workspace and Gemini Enterprise offerings. Organizations should verify the current list of covered services, editions, and features before processing PHI. Google excludes Gemini in Chrome, and personal Gemini accounts do not appear on its covered-services lists.

Much of the healthcare risk, in practice, sits between the covered and uncovered surfaces. Picture a nurse who opens Gemini on a personal Google account instead of the hospital’s covered Workspace tenant. The BAA governs how Google handles PHI inside covered services. It does not decide which surface an employee opens, what they paste, or what an AI agent does with connected records.

This article explains which Gemini products Google’s BAA covers and which obligations stay with the covered entity. It then shows how AI risk management lets healthcare teams expand Gemini use with evidence regulators can review. That work covers workforce governance, runtime defense, and agent oversight.

Key takeaways

  • HIPAA compliance depends on the Gemini deployment, not a blanket product certification: a covered service, signed Business Associate Agreement (BAA), and suitable configuration are essential. Eligible options can include specified Google Workspace and Gemini Enterprise offerings, so verify Google’s current covered-services list; personal accounts and Gemini in Chrome fall outside that coverage.
  • A Google BAA does not transfer the healthcare organization’s own PHI obligations. Unauthorized tools, minimum-necessary access, audit controls, and agents able to retrieve or send records without another human prompt still require oversight.
  • Controls applied during AI use can intercept PHI exposure rather than merely document it afterward. WitnessAI supports enterprise AI governance through AI usage discovery, context- and intent-based policies, data protection controls, and audit trails. Its agentic security capabilities extend visibility and runtime controls to supported agent workflows and tool interactions.
  • Regulatory and board scrutiny requires evidence beyond a signed agreement. AI-specific risk assessments, workforce education, continuous monitoring, and accessible audit trails help substantiate a Gemini program’s safeguards.
  • Building an AI inventory now supports existing risk-analysis duties even though the HIPAA Security Rule update remains proposed. That proposal would expressly place AI software interacting with ePHI in a written technology asset inventory.

What is Gemini HIPAA compliance?

No HIPAA certification makes an AI tool compliant on its own. HIPAA-compliant use depends on whether the service is covered by an applicable BAA and whether the covered entity or business associate meets its applicable Privacy, Security, and Breach Notification Rule obligations. The organization must also document and maintain safeguards for that use.

Google’s obligations follow from how HHS treats cloud vendors. A cloud provider that maintains ePHI for a covered entity is a business associate even if it stores only encrypted data without a key. The same HHS guidance states that a covered entity using such a provider without a BAA is in violation of the HIPAA Rules.

HHS also notes that a provider is not responsible for compliance failures attributable solely to the customer’s actions or inactions. For Gemini, the organization still owns edition selection, configuration, access scoping, and its own risk analysis.

Which Gemini products Google covers under its BAA

Google lists Gemini coverage in two places, one for Google Workspace services and one for Google Cloud services. The covered list differs by surface and, for Gemini Enterprise, by edition.

Workspace Gemini and the Gemini app

Google’s list of Workspace supported services includes the Gemini App (excluding Gemini in Google Chrome), Gemini in Google Workspace, and the Gemini Mac App. Google last updated that list on October 28, 2026.

Google’s Workspace HIPAA guidance notes that the BAA does not extend to Additional Google Services. For licensed Workspace users, Google’s Workspace with Gemini FAQ states that Gemini app submissions “aren’t reviewed by human reviewers or otherwise used to train generative AI models.” The same FAQ notes that some features may be blocked for customers who have signed the BAA, so confirm edition and feature eligibility with Google in writing.

Gemini Enterprise and Gemini models on Google Cloud

Google’s Gemini Enterprise compliance page lists HIPAA for Gemini Enterprise Standard and Plus editions and Gemini Notebook Enterprise. Business and Frontline editions are not named there. The same page limits customer-managed encryption keys to US and EU multi-region APIs.

Google describes Gemini Enterprise Agent Platform as the evolution of Vertex AI, and the same HIPAA page lists Gemini Enterprise Agent Platform and Generative AI on Agent Platform among Google Cloud supported services. Organizations should confirm the covered Google Cloud services, applicable Gemini model endpoints, features, and contractual conditions before using them to process PHI.

Gemini surfaces outside the BAA

Google blocks Gemini in Chrome for customers who have signed the HIPAA BAA, and the certifications achieved for the Gemini app do not extend to it. Audio Overviews in the Gemini app are not available to Workspace customers who have signed the BAA.

The Workspace version of Gemini Notebook does not appear on Google’s Workspace list. Google’s Gemini API terms do not mention HIPAA or BAAs, and they bar using the services in clinical practice.

Where a signed BAA stops protecting PHI

The BAA establishes contractual responsibilities for Google’s handling of PHI within covered services. Healthcare organizations remain responsible for their own use, access decisions, configurations, and safeguards, including risks arising from unauthorized AI tools and connected workflows.

Consumer accounts and shadow AI

Unsanctioned AI use is already common in hospitals. In a December 2025 survey of 518 healthcare professionals sponsored by Wolters Kluwer Health, 40% had encountered an unauthorized AI tool at work without using it, and another 17% admitted using one. Almost half of those users cited faster workflows as the reason.

Consumer Gemini handles patient data under different terms. Under Google’s Gemini Apps Privacy Notice, chats reviewed by human reviewers are retained for up to three years, even after a user deletes activity. Activity is also used to train generative AI models when Keep Activity is on.

Obligations that stay with the covered entity

Signing Google’s BAA does not eliminate the covered entity’s HIPAA responsibilities, including risk analysis, access management, audit controls, workforce safeguards, and applicable minimum-necessary requirements. The Security Rule requires audit controls that record and examine activity in systems containing ePHI, and the Privacy Rule sets a minimum necessary standard for PHI use and disclosure. None of these responsibilities transfers to Google with the BAA.

Because exposure through unsanctioned tools still lands on the covered entity, bring Legal, HR, and clinical leadership into writing and enforcing the acceptable-use policy.

Meeting the audit-control duty depends on records the covered entity can review. Google’s data governance documentation says it may log prompts to detect abuse and that some models and features carry additional prompt and response logging. That logging serves Google’s abuse detection, so confirm with Google during procurement what prompt-level records customers can access.

Agents that act on connected data

Gemini now takes actions on connected data. Google launched Gemini Enterprise Agent Platform in April 2026 and describes agents interacting across multiple systems “often without security and governance guardrails.” Depending on its permissions, integrations, and approval settings, an AI agent connected to services such as Gmail, Drive, or Calendar may be able to retrieve or transmit sensitive information without requiring approval for each action.

A February 2025 disclosure to Google described indirect prompt injection against Gemini-powered assistants, including data exfiltration scenarios, which Google has since mitigated. For agent workflows involving PHI, organizations should evaluate pre-execution authorization, least-privilege permissions, tool restrictions, and human approval for higher-risk actions.

How AI risk management makes Gemini use defensible

AI risk management treats the BAA as the foundation and adds controls between people, agents, and the model. WitnessAI is a unified AI security and governance platform, the confidence layer for enterprise AI. It enables Global 2000 organizations to observe, control, and protect AI activity across human employees and autonomous AI agents. For healthcare organizations, WitnessAI can apply supported governance and data protection controls to AI interactions routed through configured enforcement points. Coverage depends on the deployment architecture and the AI applications and workflows involved.

1. Discover where AI touches PHI

An AI inventory starts with knowing which tools staff actually use, sanctioned or not. The proposed Security Rule update would require AI software that interacts with ePHI to appear in a written technology asset inventory, though it is still a proposal. WitnessAI Observe provides AI usage discovery and visibility across supported browser-based, native, and embedded AI applications. Its AI application catalog includes more than 5,000 applications, while interaction-level visibility depends on the integration and supported traffic path. WitnessAI supports network-based deployment options that do not require endpoint agents or browser extensions. Agentic observability depends on the supported integration and the availability of agent and tool interaction telemetry.

2. Enforce intent-based policies at the point of use

Intent-based classification lets policy respond to what a clinician is trying to do, not just which keywords appear. That matters for minimum-necessary handling, because a medication history can match no DLP pattern and still be PHI. Classification in Control uses ML models that analyze conversational context and purpose. Depending on policy configuration and deployment, WitnessAI can allow, warn, block, or route supported AI interactions to approved models. This gives organizations alternatives to blanket restrictions while maintaining control over sensitive AI use.

3. Tokenize PHI before it reaches a model

Data tokenization can reduce the exposure of detected patient identifiers by replacing sensitive values before they are sent to a model. OWASP’s guidance on LLM data exposure names health records as sensitive information and recommends tokenization to sanitize it before processing. WitnessAI supports real-time tokenization of detected sensitive information in supported AI workflows, reducing exposure to external models. The specific tokenization and data-restoration behavior should be confirmed for the configured deployment.

4. Govern agents before they act

Agent governance has to act before a tool call runs, not after it. Witness Protect provides runtime security controls for supported AI models and applications, including prompt injection and jailbreak detection and response filtering. Agentic runtime controls address supported tool interactions and actions, extending protection beyond model inputs and outputs. WitnessAI’s agentic security capabilities support policy-based governance of agent tool interactions at supported enforcement points, including controls over permitted tools and actions. Where identity and execution context are available, agent activity can be associated with the initiating user or relevant agent identity, supporting investigation and accountability.

5. Keep audit trails regulators can review

Audit trails for AI interactions captured through WitnessAI give compliance teams a reviewable record of prompts and responses routed through the platform. Those audit trails are immutable, and the data can be exported to SIEM tools for unified visibility. WitnessAI’s enterprise architecture includes single-tenant deployment and bring-your-own-key (BYOK) capabilities, with SOC 2 Type II compliance documented in its enterprise messaging. These controls support enterprise security and data governance requirements, subject to deployment configuration.

WitnessAI secures AI use for more than 820,000 workers globally, monitors millions of AI interactions daily, and reports 99.3% true-positive guardrail efficacy.

A Global Top 5 Airline’s VP of Cybersecurity: “The ability to see every AI interaction across our global workforce has transformed our security posture.”

WitnessAI for Employees
FOR EMPlOYEES

Your Employees Are Already Using AI. Are You Governing It?

WitnessAI gives you full visibility into employee AI usage, classifies intent behind every interaction, and enforces smart policies, without slowing anyone down.

Learn About WitnessAI For Employees

Proving Gemini HIPAA compliance to regulators and boards

Regulators and accreditors look for evidence that AI safeguards work in practice, not only a signed agreement. The Joint Commission’s responsible AI guidance, developed with the Coalition for Health AI, calls for data security protections, ongoing quality monitoring, and staff education. Google’s BAA addresses its contractual responsibilities for covered services. Healthcare organizations also need appropriate administrative, physical, and technical safeguards. AI visibility, intent-based controls, and audit trails can contribute to those safeguards, but do not independently establish HIPAA compliance.

WitnessAI can provide governance records and audit evidence for supported employee and agent AI activity. These records can contribute to internal reviews, risk assessments, and board reporting as organizations evaluate broader Gemini deployment. To see how it fits your Gemini environment, schedule a demo.

FAQs about Google Gemini HIPAA compliance

Gemini’s HIPAA eligibility depends on the service, account, and BAA. These answers distinguish covered deployments from consumer surfaces and explain the proposed Security Rule update’s implications.