HIPAA-Compliant AI Agent Platforms: 15 No-Code Builders Checked for a BAA (2026)

Part of our complete guide to HIPAA requirements for healthcare software.

In brief: HIPAA compliant AI agents need three things: a signed BAA with every vendor that touches PHI, access limited to the minimum PHI each task needs, and a log of everything the agent reads and does. A HIPAA-compliant AI agent platform needs a signed BAA that covers the plan and features you use, plus safeguards you configure and test: access control, audit logs and encryption. Of 15 platforms rechecked on 30 October 2026, eight offer a BAA, three are not for PHI, two need self-hosting and two are unconfirmed.

The eight that offer a BAA are UiPath, Microsoft Power Automate and Copilot Studio, Voiceflow and Lindy (Enterprise plans), Keragon (all paid plans), and Google Cloud's and AWS's agent platforms under their cloud BAAs. Zapier, Make and Relevance AI are not for PHI, n8n and Flowise are usable only when self-hosted on BAA-covered infrastructure, and Magical and Botpress claim compliance without a published BAA.

Updated October 2026. Vendor status was rechecked on the vendor's own pages on 30 October 2026; a vendor can change its terms, so get the signed BAA before PHI flows. Need one built instead? See our HIPAA compliant AI agent development service.

Which AI agent platforms are HIPAA compliant? The 15-platform matrix

Eight of the 15 platforms offer a BAA on at least one plan. Every row below was checked on the vendor's own pages on 30 October 2026, except the UiPath audit column, last checked on 24 October. "HIPAA compliant" on a marketing page is not the same as a signed BAA, and HHS does not certify any product as HIPAA compliant, so the BAA column is the first gate: without a BAA, PHI cannot flow through a platform at all. After that, the plan, the covered features and the safeguards you configure decide whether a specific use is allowed.

Key Takeaways from the Matrix

Eight platforms sign a BAA: UiPath, Microsoft Power Automate and Copilot Studio, Voiceflow and Lindy on Enterprise plans, Keragon on every paid plan, and the Google Cloud and AWS agent platforms under their cloud BAAs.

Two claim HIPAA compliance without a published BAA: Magical and Botpress. Ask for a signed BAA before any PHI touches them.

Two can handle PHI only when you self-host them on infrastructure already covered by a BAA: n8n and Flowise. Your team then owns encryption, access control, logging and patching.

Three are not for PHI: Zapier says so directly, Make publishes no BAA, and Relevance AI publishes no HIPAA terms. Use them only for workflows that never touch patient data.

HIPAA compliant AI tools: the BAA covered features to confirm

HIPAA compliant AI tools are only compliant for the services their BAA lists, so confirm coverage feature by feature before PHI reaches them. A BAA can cover one plan and not another, one region and not another, and generally available services but not preview features; Google Cloud's BAA, for example, excludes pre-release features.

• Plan: Enterprise-only BAAs (Voiceflow, Lindy) do not extend to lower tiers.
• Region: some plans are hosted outside the US.
• Model: the model provider behind the agent needs its own BAA with the platform, as a subprocessor.
• Connectors: each third-party app the agent calls is another place PHI goes.
• Beta features: preview tools, including computer use, are often outside the covered-services list.

Who this guide is for

If you are choosing an AI agent platform for patient data, you are usually a CIO, CTO, compliance lead or product head deciding one thing: can this platform touch PHI under a signed BAA, for the plan and features we will actually use? The fears behind that question are concrete: a vendor that says "HIPAA compliant" but will not sign, a model provider missing from the BAA chain, an agent that writes to the wrong patient record, and a pilot that turns into a maintenance bill. This guide answers the first question for 15 platforms, then shows what to check before an agent acts on PHI. If no platform fits your workflow, our team builds custom HIPAA-compliant AI agents for healthcare on infrastructure you already cover with a BAA.

Evaluating agents for prior authorization specifically? Our AI prior authorization guide covers the compliance sign-off and where a human must stay in the loop. For a payer-specific purchase, use our prior authorization software buyer's guide.

No-code AI promises agents in days instead of months, built by the people who know the workflow, and a pilot before anyone commits to a custom engineering project. For healthcare organizations, this promise is especially compelling. Pressure to adopt AI for operational efficiency, patient engagement and revenue cycle work is coming from every direction: boards, payers and patients.

But here is the problem that generic "best no-code AI builder" lists ignore entirely: most no-code AI platforms cannot legally handle Protected Health Information (PHI). Without a Business Associate Agreement (BAA), a covered entity may not send PHI to the platform at all; encryption, audit logging and an independent audit report such as SOC 2 Type II are what your security team checks next. Deploying a no-code AI agent on patient data without a BAA violates federal HIPAA rules. HIPAA civil penalties, as last adjusted by HHS effective 28 January 2026, run from $145 to $73,011 per violation for lower tiers and up to $2,190,294 per violation for uncorrected willful neglect, with an annual cap of $2,190,294 per identical provision.

If compliance itself is the bottleneck, our healthcare compliance software services cover HIPAA risk analysis, safeguards and audit readiness end to end.

This guide checks 15 AI agent builders, from no-code tools to the cloud agent platforms, against healthcare-specific requirements. For each platform, we assess BAA availability, SOC 2 report status, PHI handling capabilities, healthcare use cases, limitations, and pricing. If you are a healthcare CIO, IT director, compliance officer, or innovation lead evaluating no-code AI tools, this comparison tells you which platforms you can actually use with patient data and which ones will put your organization at risk. The list mixes three kinds of product: no-code builders (Zapier, Make, Magical, Lindy, Keragon, Voiceflow), low-code and open-source tools (UiPath, Power Automate, Copilot Studio, n8n, Flowise, Botpress, Relevance AI), and developer cloud platforms (Google Vertex AI Agent Builder, AWS Bedrock AgentCore).

Why Healthcare Teams Want No-Code AI Agents

Healthcare teams want no-code AI agents because the people who know the workflow best, revenue cycle managers, patient access leads and clinical operations staff, can build and test an agent themselves instead of waiting for engineering. Low-code tools also let IT deploy agents across hospital departments from one governed platform, and a pilot on synthetic data produces the evidence a budget request needs.

A denial management supervisor who knows which CARC and RARC code patterns signal a preventable denial can turn that knowledge into an agent that flags those claims automatically, and an operations lead can pilot AI-driven prior authorization triage in weeks. The catch is the contract and the controls, which is what the rest of this guide checks. As we explain in our guide on building HIPAA-compliant AI agents, deciding the architecture early prevents rework later, even on a no-code platform.

HIPAA Requirements for AI Platforms: The Non-Negotiable Checklist

A HIPAA compliant AI platform must sign a BAA, encrypt ePHI, log access and limit each user and agent to the minimum PHI it needs; most security teams also ask for a SOC 2 Type II report or HITRUST certification. Before evaluating any no-code AI platform for healthcare use, you need to understand these four HIPAA requirements, plus the audit evidence your security team will ask for, that apply to any technology that creates, receives, maintains, or transmits electronic Protected Health Information (ePHI). The HHS HIPAA Security Rule mandates specific administrative, physical, and technical safeguards that every covered entity and business associate must implement.

1. Business Associate Agreement (BAA)

Any vendor that handles PHI on behalf of a covered entity is a business associate under HIPAA. The BAA is a legally binding contract that specifies how the vendor will protect PHI, report breaches, and comply with HIPAA obligations. Without a signed BAA, a covered entity may not disclose PHI to the vendor, so the platform cannot store, process or transmit it. This is the single most important criterion. No BAA means the platform is disqualified for any workflow involving patient data, regardless of how good its security features are.

2. Encryption at Rest and In Transit

ePHI must be encrypted both when stored (at rest) and when transmitted (in transit). Industry standards require AES-256 for storage encryption and TLS 1.2 or higher for data in transit. Some platforms encrypt data at the application layer, while others rely on infrastructure-level encryption from cloud providers like AWS or Azure. Both approaches can satisfy HIPAA requirements, but the covered entity must verify the specifics.

3. Audit Logging

HIPAA requires covered entities and business associates to maintain logs of all access to ePHI, including who accessed what data, when, from where, and what actions they performed. HIPAA's audit controls standard (45 CFR 164.312(b)) sets no retention period for the logs themselves; the six-year rule in 164.316(b)(2) covers policies, procedures and records of required activities. Most organizations keep audit logs for six years anyway, make them tamper-evident, and keep them available for audits and breach investigations. For AI platforms, this means logging every interaction between the AI agent and any PHI-containing system or data field.

4. Access Controls and Role-Based Permissions

The principle of least privilege applies: users and AI agents should only have access to the minimum PHI necessary to perform their function. This requires role-based access control (RBAC), multi-factor authentication (MFA), automatic session timeouts, and unique user identification for audit trail purposes. For no-code platforms where multiple team members may be building and editing agents, granular access controls become especially important.

5. Independent Audit Evidence: SOC 2 Type II or HITRUST

HIPAA does not require either one. A SOC 2 Type II report (an independent auditor's attestation, not a certification) and HITRUST r2 certification are the evidence most security teams ask for that a vendor operates the controls HIPAA expects. A SOC 2 Type II audit evaluates controls over an extended period (typically 6 to 12 months), providing assurance that security is operational, not just documented. HITRUST CSF is specifically designed for healthcare and maps directly to HIPAA requirements. Vendors with neither require significantly more due diligence from your compliance team. Additional compliance certifications such as ISO 27001 help, though they do not replace a SOC 2 Type II report or HITRUST for most enterprise security reviews.

15 AI Agent Builders Checked for HIPAA and a BAA

Eight of the 15 platforms offer a BAA on at least one plan; each profile below gives the HIPAA status, fit, limits, price and our verdict. We evaluated each platform against the requirements above, plus healthcare-specific capabilities like EHR integrations, healthcare workflow templates, and healthcare customer references. Each platform was checked on the vendor's own pages on 30 October 2026. For deeper architectural context on what healthcare AI agents require, see our analysis of 5 healthcare workflows that agentic AI will transform.

1. Magical

What it does: Magical is a Chrome extension and desktop agent that uses RPA-style automation to move data between browser tabs, auto-fill forms, and execute repetitive tasks. It is popular in healthcare revenue cycle settings for tasks like transferring patient data between EHR screens and payer portals.

HIPAA status: Magical's trust page says it is HIPAA compliant and SOC 2 Type II. The page does not mention a Business Associate Agreement, so treat BAA availability as unconfirmed and ask for one in writing before any PHI flows through it.

Healthcare use cases: Prior authorization form filling, eligibility verification data transfer, claims status checking, patient charting data entry, payment posting across systems.

Limitations: Magical operates as a browser extension, which introduces unique security considerations. Browser extensions have broad access to page content and could theoretically capture PHI displayed on screen. The RPA approach is inherently fragile, breaking when payer portals or EHR interfaces change their layout. It is a task automation tool, not an AI agent builder in the traditional sense; it cannot reason, make decisions, or handle complex multi-step workflows with branching logic.

Pricing: Pricing is not published; contact sales.

HIPAA verdict: Unconfirmed. A "HIPAA compliant" claim is not a BAA. Use it with PHI only after Magical signs one, and govern the browser extension tightly.

2. UiPath

What it does: UiPath is an enterprise RPA platform offering both attended and unattended robotic process automation, AI document understanding, process mining, and increasingly, AI-powered agent capabilities through its Automation Cloud. UiPath has a dedicated healthcare vertical with pre-built automation templates.

HIPAA status: UiPath publishes a Business Associate Agreement covering the products listed on its trust portal, holds a SOC 2 Type 2 report, and announced HITRUST r2 certification for Automation Cloud in August 2023 (an r2 certificate runs two years, so ask for the current one). Enterprise customers can deploy on-premises.

Where it fits in healthcare: Claims processing automation, prior authorization workflows, patient scheduling optimization, clinical documentation extraction, revenue cycle management automation, insurance eligibility verification, medical records requests, coding and billing automation.

Limitations: UiPath is complex. The learning curve for building sophisticated healthcare automations is significant, often requiring dedicated UiPath developers (a "citizen developer" can handle simple automations, but complex healthcare workflows need expertise). Pricing is enterprise-grade, which puts it out of reach for small practices and clinics. It is primarily RPA-focused; while AI capabilities are growing, it is not a native AI agent builder in the way that LLM-based platforms are.

Pricing: Automation Cloud Basic starts at $25 a month but is hosted in the EU only; US healthcare buyers will be on Standard or Enterprise, which are priced by sales.

HIPAA verdict: Signs a BAA. The strongest enterprise option, with on-premises deployment available, and matching enterprise complexity and cost.

3. Relevance AI

What it does: Relevance AI is an API-first AI agent builder that lets teams create custom AI agents using a visual workflow builder with drag-and-drop interface. The platform supports memory, variables, vector databases, and integrates with tools like Slack, Google Workspace, HubSpot, and Notion. It is designed for business teams automating internal operations.

HIPAA status: Relevance AI's security documentation confirms SOC 2 Type II and a choice of AU, US or EU/UK data residency. Its data-retention docs say retention settings can support frameworks such as HIPAA, but Relevance AI publishes no BAA, so treat it as not for PHI until it signs one.

Healthcare applications: Internal operations automation (non-PHI), patient feedback analysis (de-identified data only), provider credentialing workflow support, administrative task automation.

Limitations: Without a confirmed BAA, Relevance AI cannot be used for any workflow that touches PHI. It is not designed with healthcare-specific features; there are no EHR integrations, no healthcare workflow templates, and no published healthcare customer case studies. The API-first architecture, while flexible, requires more technical skill than truly no-code drag-and-drop tools.

Pricing: Free tier and paid plans; enterprise pricing on request.

HIPAA verdict: No BAA published. Keep it to workflows with no PHI or with de-identified data only.

4. Voiceflow

What it does: Voiceflow is a conversational AI platform that lets teams build voice and chat AI agents using a drag-and-drop editor. It is popular for customer service chatbots, interactive voice response (IVR) systems, and conversational interfaces across industries.

HIPAA status: Voiceflow's healthcare page states that "a BAA is available on enterprise plans" and that the platform is SOC 2 Type II audited. Its HIPAA page asks customers to contact its security team about the BAA for their use.

Typical healthcare workflows: Patient-facing voice and chat agents for scheduling, reminders, FAQs and intake; with the Enterprise BAA in place these can handle patient identifiers, and without it they must stay free of PHI.

Limitations: The BAA is Enterprise-only, so smaller teams on lower tiers cannot use it with PHI. EHR integration is custom work through APIs, and you still own the design of consent, identity checks and escalation to staff.

Pricing: Enterprise pricing is on request; lower tiers do not include a BAA.

HIPAA verdict: Signs a BAA on Enterprise plans. A realistic choice for patient-facing voice and chat agents once the Enterprise BAA is in place.

5. Botpress

What it does: Botpress is an open-source chatbot and conversational AI platform. Originally designed for developers, it has added visual flow builders and low-code capabilities. The key differentiator for healthcare is that Botpress can be self-hosted, giving organizations full control over data handling.

HIPAA status: A Botpress blog post says it offers "compliance with" HIPAA, and its trust center lists HIPAA and a SOC 2 Type 2 report, but neither publishes BAA terms; the documents sit behind an access request. Its self-hosted versions, including v12, are officially sunset, so self-hosting is no longer a route to compliance.

Healthcare use cases (only once a BAA is signed): Patient intake chatbots, symptom triage conversations, appointment scheduling with patient identification, medication reminder bots, post-discharge follow-up conversations, clinical trial screening questionnaires.

Limitations: With self-hosted versions sunset, Botpress now runs as a cloud service, so PHI use depends entirely on the BAA and security terms Botpress will sign. Get both in writing first.

Pricing: Plus $150 and Team $750 a month, billed annually; Enterprise priced on request.

HIPAA verdict: Unconfirmed. Ask for a signed BAA and the current SOC 2 report before using it with PHI.

6. Zapier

What it does: Zapier is the most widely used workflow automation platform, connecting 9,000+ apps through trigger-action "Zaps." It has added AI features including natural language automation creation and AI-powered data transformation. Zapier is ubiquitous in business operations for connecting SaaS tools.

HIPAA status: Zapier says it plainly: "Zapier isn't HIPAA compliant" and customers should not use it to automate anything involving PHI (Zapier). It does not sign a BAA on any plan.

Use cases: Strictly non-PHI workflows only: marketing automation for healthcare organizations, staff scheduling notifications, facility management alerts, vendor management workflows, financial reporting aggregation (without patient-level data).

Limitations: The prohibition on PHI handling eliminates the majority of healthcare automation use cases. Any Zap that touches patient records, billing data, clinical notes, insurance information, or appointment details involving patient identifiers is off-limits. Even apparently harmless automations like "when a new patient books an appointment, send a confirmation" involve PHI (patient name + appointment details) and cannot use Zapier.

Pricing: Free and paid plans; no plan includes a BAA.

HIPAA verdict: Not for PHI. Fine for marketing, HR and facilities workflows that never touch patient data.

7. Make (formerly Integromat)

What it does: Make is a visual workflow automation platform similar to Zapier but with more complex branching logic, data transformation, and multi-step scenario capabilities. Its visual builder uses a node-based canvas that many users find more intuitive for complex workflows than Zapier's linear approach.

HIPAA status: Make's security page lists SOC 2 Type II, SOC 3 and ISO 27001 but says nothing about HIPAA or a BAA, and its community forum answers say it does not sign one. Its on-premises agents only bridge to local systems; the platform itself stays in Make's cloud.

Where healthcare teams use it: Non-PHI operations only: marketing campaign automation, social media management, vendor invoice processing, facility maintenance request routing, HR onboarding workflows for healthcare staff.

Limitations: Same fundamental limitation as Zapier: no PHI handling. The EU regulatory focus means HIPAA is not on the product roadmap. Healthcare organizations using Make for operational workflows must implement strict controls to ensure no PHI ever enters a Make scenario, which requires ongoing training and monitoring.

Pricing: Free plan, paid plans from $9 a month, Enterprise on request; none includes a BAA.

HIPAA verdict: No BAA. Keep PHI out.

8. Microsoft Power Automate

What it does: Power Automate (formerly Microsoft Flow) is Microsoft's enterprise workflow automation platform, part of the Power Platform alongside Power BI, Power Apps and Copilot Studio. It offers cloud flows, desktop flows (RPA), and AI Builder for intelligent automation. Deep integration with Microsoft 365, Azure, Dynamics 365, and third-party connectors.

HIPAA status: Power Automate is covered by Microsoft's HIPAA BAA, which Microsoft includes by default for covered entities and business associates. Microsoft's BAA-covered services are audited for HITRUST CSF and ISO 27001.

Healthcare scenarios: Patient appointment reminders and scheduling, clinical documentation workflows, claims processing and follow-up automation, care coordination notifications, referral management, lab result routing, medication refill request processing, patient intake form processing with AI Builder, integration with Epic and Cerner through Azure FHIR service.

Limitations: Power Automate's strengths are also its constraints. It is most powerful within the Microsoft ecosystem; organizations not already invested in Microsoft 365 and Azure face a steep onboarding curve. AI capabilities through AI Builder are more limited than dedicated AI agent builders. Complex healthcare AI agents (multi-turn reasoning, clinical decision support, dynamic care pathway navigation) exceed what Power Automate can natively do. Licensing complexity is a real issue: healthcare organizations must carefully audit which licenses and services are covered by the HIPAA BAA.

Pricing: Power Automate Premium is listed at $15 per user per month, paid yearly; process and hosted RPA capacity are priced separately.

HIPAA verdict: Signs a BAA by default. The easiest compliant choice for organizations already on Microsoft 365 and Azure.

9. n8n

What it does: n8n is an open-source workflow automation platform often described as a "self-hostable Zapier." It provides a visual node-based editor for building complex automation workflows, includes AI agent capabilities with LLM integrations, and offers both cloud and self-hosted deployment options.

HIPAA status: n8n holds SOC 2 Type 2 and SOC 3 reports (trust center), but we found no BAA on its site, trust center or legal pages, so n8n Cloud should not process PHI. The route for PHI is self-hosting n8n on infrastructure you already cover with a BAA, such as AWS, Azure or Google Cloud, where your team owns every control.

Healthcare use cases (self-hosted): EHR data extraction and transformation, patient appointment workflow automation, clinical document processing with AI, insurance verification automation, medical records request routing, FHIR API integration workflows, discharge follow-up automation, lab result notification pipelines.

Limitations: Self-hosting requires significant DevOps capability. You need to configure encryption at rest and in transit, implement audit logging, enforce RBAC with MFA, conduct regular penetration testing, and manage ongoing security patching. The n8n community has published guides for HIPAA-compliant deployment on AWS EKS, but the responsibility and effort are substantial. AI agent capabilities, while powerful, require connecting external LLM providers that must also be HIPAA compliant (Azure OpenAI with BAA, AWS Bedrock with BAA).

Pricing: n8n Cloud has Starter, Pro, Business and Enterprise plans; the Community Edition is free to self-host.

HIPAA verdict: No BAA from n8n. Usable for PHI only when self-hosted on BAA-covered infrastructure with your own security controls.

10. Flowise

What it does: Flowise is an open-source AI agent builder powered by LangChain that transforms complex AI development into a visual drag-and-drop experience. Unlike workflow automation tools (Zapier, Make, n8n), Flowise is specifically designed for building AI agents with LLM integration, RAG (Retrieval-Augmented Generation) pipelines, and conversational AI applications. Its source is on GitHub.

HIPAA status: Flowise is open source and was acquired by Workday in August 2025. Its own pages say nothing about HIPAA, SOC 2 or a BAA; claims elsewhere that it is "HIPAA compliant" come from automated profiles, not from Flowise.

Healthcare use cases (self-hosted or Enterprise): Clinical documentation AI assistants, patient-facing health information chatbots with RAG over medical knowledge bases, medical records summarization, prior authorization support agents, clinical trial matching, provider directory search agents, drug interaction checking agents.

Limitations: Flowise is the most specialized AI agent builder in this list, which means it lacks the broad workflow automation capabilities of tools like n8n or Power Automate. It excels at LLM-powered agents but is not designed for general business process automation like form filling, data transfer between systems, or RPA-style browser automation. Self-hosting requires familiarity with Node.js deployment and the LangChain ecosystem. Healthcare organizations need to ensure that the underlying LLM provider (OpenAI, Anthropic, Azure) also has HIPAA coverage.

Pricing: Open source and free to self-host; Flowise Cloud Starter $35 a month and Pro $65 a month, with no BAA.

HIPAA verdict: No BAA. Usable for PHI only when self-hosted on BAA-covered infrastructure, with your team responsible for every control.

11. Microsoft Copilot Studio

What it does: Microsoft's agent builder for chat and action agents across Microsoft 365, Teams and custom channels.

HIPAA status: Copilot Studio is listed as covered under the Microsoft HIPAA BAA, inheriting Microsoft's HITRUST, SOC and ISO programs.

Pricing: $200 per month for a pack of 25,000 Copilot Credits, or pay-as-you-go through an Azure subscription.

HIPAA verdict: Signs a BAA. The natural agent builder for organizations already on Power Automate and Microsoft 365.

12. Lindy

What it does: A no-code builder for AI assistants that handle email, scheduling, calls and follow-ups.

HIPAA status: Lindy offers a BAA on its Enterprise plan and holds a SOC 2 Type II report; its Free and Team plans do not include a BAA.

Pricing: Enterprise pricing on request; the Team plan starts at $49.99 a month without a BAA.

HIPAA verdict: Signs a BAA on Enterprise only.

13. Keragon

What it does: A healthcare-specific no-code automation platform with prebuilt connectors to EHRs and practice systems.

HIPAA status: Keragon includes a BAA on all paid plans and holds a SOC 2 Type II report.

Pricing: Starter is $149 a month billed monthly, or $99 a month billed annually, and paid plans include a BAA.

HIPAA verdict: Signs a BAA on every paid plan. The most accessible healthcare-first option for smaller teams.

14. Google Vertex AI Agent Builder (Gemini Enterprise Agent Platform)

What it does: Google Cloud's platform for building and deploying AI agents on Gemini models.

HIPAA status: Google Cloud's HIPAA BAA covers the agent platform, with two limits to respect: Google says not to use pre-GA offerings with PHI, and not to put PHI in Conversational Agents definitions, including intents, training phrases and entities.

Pricing: Pay-as-you-go on Google Cloud.

HIPAA verdict: Signs a BAA (self-serve through Google Cloud). Developer-oriented rather than truly no-code.

15. AWS Bedrock AgentCore

What it does: AWS's managed runtime for building and operating AI agents on Bedrock models.

HIPAA status: Bedrock AgentCore appears on AWS's HIPAA-eligible services list, so it is covered by the AWS BAA once your account has one.

Pricing: Pay-as-you-go on AWS.

HIPAA verdict: Covered by the AWS BAA. Developer-oriented; see our guide to building HIPAA-compliant AI agents for the architecture.

One more worth knowing: Retool's subscription agreement bars PHI on Retool Cloud, so only self-hosted Retool on your own BAA-covered infrastructure can touch patient data.

Healthcare-Specific Autonomous AI Agents: What HIPAA Adds When Agents Act

Healthcare-specific autonomous AI agents are HIPAA compliant only when the BAA covers every service they call and the agent itself is limited to the minimum PHI and the actions its task needs. A chatbot that answers questions reads PHI; an autonomous agent also acts on it, by booking, updating, submitting or messaging, so the controls have to cover its actions as well as what it reads.

• Minimum necessary applies to the agent. HIPAA's minimum necessary standard (45 CFR 164.502(b)) limits uses and disclosures to what the task needs. Give each agent its own service identity scoped to the fields and records of its workflow, never a shared staff login with full access.
• Actions need an allowlist and a human gate. List the actions an agent may take. Reads can run unattended; writes to a chart, a claim or a patient message should pass an identity check (two identifiers match) and, for anything clinical or financial, a human approval until test results justify removing it.
• Agents that operate screens carry extra PHI risk. Computer-use agents, which read the screen and click through an EHR or payer portal when no API exists, send screenshots to a model. Those screenshots contain PHI, so the model endpoint, the screenshot store and every log must sit under a BAA, and screen automation must be permitted by the EHR or portal's terms.
• Model terms decide your exposure. Confirm in writing that prompts and outputs are not used for training and how long they are retained, for the platform and for each model provider behind it.
• Prompt injection is a security risk. Text inside a referral, an email or a web page can instruct an agent to do something else. OWASP lists prompt injection as the first risk for LLM applications; scoped permissions and action allowlists are what stop an injected instruction from reaching PHI.
• Every action needs a trace. Log who or what triggered the agent, which records it read, which action it took, what it checked before acting, and who approved it. That trace is what your compliance team needs after an incident, and it supports the audit controls standard (45 CFR 164.312(b)).

HIPAA-Compliant AI Agents by Use Case

What a BAA Covers, and What It Does Not

A BAA is a contract in which a vendor agrees to protect PHI the way HIPAA requires; it does not by itself make your agent compliant. Under the HIPAA Privacy Rule (45 CFR 164.502(e) and 164.504(e)), a covered entity may share PHI with a business associate only after a written agreement that limits how the vendor uses the data, requires safeguards, and requires breach reporting. Three points catch AI agent buyers most often:

• The chain has to be unbroken. If the agent platform sends PHI to a model provider, a speech-to-text service or a vector database, each of those is a subcontractor, and the platform needs its own BAA with each one (45 CFR 164.502(e)(1)(ii)). Ask the platform for its list of subprocessors that see PHI.
• "We do not store data" is not an exemption. HHS guidance on cloud computing says a service that creates, receives, maintains or transmits ePHI is a business associate even if it only processes the data in transit or keeps it encrypted. The "conduit" exception is narrow and meant for services like internet providers and couriers.
• A BAA covers only the products it names. Many vendors sign a BAA for an Enterprise plan or a list of eligible services. Using PHI on a plan or feature outside that list is not covered, even under the same account.

The BAA is the first check, not the last one. You still need your own risk analysis, access controls, audit logs and a decision about where a human must review the agent's output.

Minimum necessary: how much PHI an AI agent may see

An AI agent may reach only the PHI its task needs, because HIPAA's minimum necessary standard applies to its uses and disclosures just as it does to staff. The rule sits at 45 CFR 164.502(b), and 45 CFR 164.514(d) tells you to define which roles need which PHI, so treat each agent as its own role.

The easy mistake is to connect an agent with a staff login or an API key that can read the whole chart. A model that can read everything will sometimes use everything, in its reasoning, its output and its logs. The standard has a narrow exception for disclosures to a provider for treatment, but it does not cover a scheduling or billing agent doing operations work.

These rows are starting points. Your privacy officer sets the final scope. Enforce it in the integration layer, with scoped FHIR queries or a filtered API, rather than by telling the model to ignore fields in its prompt, because a prompt instruction is not an access control.

Audit logs for AI agents: what to record

An AI agent's audit log should let you rebuild who or what touched which PHI, what the agent did with it, and who approved it. HIPAA's audit controls standard, 45 CFR 164.312(b), requires a way to record and examine activity in systems holding ePHI, but it does not list fields, so the list below is our recommendation.

Store record IDs in the log where you can, not PHI itself. Logs copied into an analytics or error-tracking tool without a BAA are a common route for PHI to leak. Restrict who can read the logs and make them tamper-evident. Most organizations keep them for six years, as noted above, though the audit controls standard sets no period.

The HIPAA breach clock when an AI agent exposes PHI

You have at most 60 calendar days from discovering a breach to notify the people affected, and "without unreasonable delay" comes first, so 60 days is an outer limit. Discovery is the first day the breach is known, or would have been known with reasonable diligence, to anyone in your workforce or any agent of yours other than the person who committed it (45 CFR 164.404(a)(2)).

Three points matter for agents:

• Your clock can start at the vendor. If the vendor acts as your agent under federal agency law, its knowledge counts as yours, so your 60 days can start when the vendor finds out. Ask counsel how your contract reads.
• Shorten the vendor's deadline in the BAA. Sixty days from the vendor leaves you no time. Negotiate a notice period in days, and name who receives the notice.
• Agent logs decide the size of the breach. Without records of what the agent read, you may have to treat every patient it could reach as affected. An impermissible disclosure is presumed to be a breach unless your risk assessment shows a low probability of compromise (45 CFR 164.402).

State breach laws can set shorter deadlines. HHS proposed a stricter Security Rule in January 2025; it was not final when we checked in October 2026, and secondary reports cite a July 2027 target, so confirm the status before you plan around it.

How to Evaluate a HIPAA-Compliant AI Agent Platform

The safest way for a healthcare startup to run agentic AI is on a platform that signs a BAA, or self-hosted on infrastructure already covered by its cloud provider's BAA; either way, budget the security engineering and hosting costs, not just the license. Evaluate an AI agent platform for a regulated healthcare company in the same order every time: the contract, then security and governance, then how agents are tested before they touch patient data.

If the platform cannot finish an EHR write, evaluate EHR automation without API access before adding a screen agent. Run this evaluation in a sandbox with synthetic patients first; our synthetic health data guide shows how to generate realistic FHIR test data.

When No Platform Covers Your Workflow: Custom HIPAA-Compliant AI Agent Development

Choose a custom HIPAA-compliant AI agent when your workflow must read or write EHR data, cross several systems, or run inside your own cloud account under your own BAA. If one of the BAA-signing platforms above covers your workflow, start there: it is faster and cheaper to begin. Nirmitee is not a no-code platform, so it is not in the matrix; we are a healthcare engineering team that builds custom AI agents on BAA-covered infrastructure such as AWS or Google Cloud and connects them to EHRs through FHIR and HL7 as part of our EHR integration services. We are ISO 27001:2022 certified and we sign BAAs.

The objections your security and operations teams will raise

How we do it

1. Scope one workflow. We map each step, the systems it touches and every place PHI moves, then list the BAAs that chain needs.
2. Pick the integration route per step. FHIR, HL7 v2 or the vendor's API first. Screen automation only where no API supports the action and the EHR or portal terms allow it.
3. Build inside your cloud. The agent runs in your AWS, Azure or Google Cloud account under your BAA, with its own scoped identity and an allowlist of actions.
4. Test the failures before go-live. On synthetic patients we test wrong-patient cases, duplicate submissions, timeouts, expired sessions and changed screens. The acceptance bar for unauthorized writes is zero.
5. Pilot with review, then widen. Staff approve the agent's actions at first; we measure completion, rework and review time, and remove approval steps only where the numbers support it.

The mistakes to check for first: a BAA that covers the plan but not the feature in use (Google, for example, excludes PHI in Conversational Agents definitions), a model or speech provider missing from the subcontractor chain, and agent logs with PHI flowing into an analytics or error-tracking tool that has no BAA.

Work out your own return

Use your own volumes. A simple estimate of the monthly value of an agent on one workflow is:

Monthly value = tasks per month x share completed by the agent x (manual minutes per task - review minutes per task) / 60 x loaded hourly cost - monthly platform, cloud and support cost

If you do not have a wage figure, the US median pay for medical records specialists was $24.59 an hour in 2025 (BLS), and wages were 70.0% of private employer compensation cost in June 2026 (BLS), which gives a loaded cost of about $35.13 an hour.

An illustration with sample inputs: 2,000 tasks a month, 60% completed by the agent, 6 manual minutes and 1 review minute per task gives 2,000 x 0.6 x 5 / 60 = 100 staff hours, or about $3,513 a month before platform and support costs. Replace each input with your own numbers; the share completed is the one to measure in a pilot rather than assume.

When to Use No-Code vs. Custom Development

Use no-code when the workflow is well defined, stays inside the platform's connectors and the vendor signs a BAA for your plan; build custom when the agent must write to an EHR, cross several systems or run in your own cloud. The decision depends on the complexity of the workflow, the sensitivity of the data, the integration requirements, and the organization's long-term AI strategy. For organizations considering the custom route, our healthcare AI solutions page details how custom architectures address the limitations that no-code platforms cannot.

Choose No-Code When:

• The workflow is well-defined and repetitive: Prior authorization status checks, eligibility verification, claims follow-up, appointment reminders. These are automation candidates, not AI agent candidates.
• You are piloting a concept: Before committing to custom development, a no-code prototype proves the value proposition and generates the data needed for a business case.
• The data does not involve PHI: Marketing automation, staff communication, vendor management, and facility operations can safely use non-HIPAA platforms like Zapier or Make.
• You need results in weeks, not months: If the use case has a time-sensitive business need (a new payer contract requiring specific claim formatting, a regulatory reporting deadline), no-code delivers faster.
• A HIPAA-compliant platform supports your use case: If Power Automate or UiPath can handle the workflow and you are already in their ecosystem, you avoid adding a new vendor to the compliance review.

Choose Custom Development When:

• The AI agent needs clinical reasoning: Diagnostic support, treatment recommendation, clinical risk scoring, and care pathway navigation require purpose-built AI models, fine-tuned on clinical data, with explainability requirements that no-code platforms cannot satisfy.
• Deep EHR integration is required: If the agent needs to read and write to Epic, Cerner, or other EHR systems through FHIR R4 APIs, handle CDS Hooks, process SMART on FHIR authorization flows, and maintain persistent patient context across sessions, no-code platforms lack the integration depth.
• You need full audit trail control: Clinical decision support systems require comprehensive audit trails that document not just what data was accessed but what reasoning the AI applied and what recommendation it generated. Check whether the platform's logs record this before you rely on them.
• Regulatory requirements demand it: FDA Software as a Medical Device (SaMD) classification, Clinical Decision Support (CDS) criteria under 21st Century Cures Act, and state-specific telehealth regulations may require documentation, testing and validation that a no-code platform may not support, so check this before you pick one.
• Scale and performance matter: A no-code platform handling 100 interactions per day is fine. An AI agent processing thousands of clinical documents, managing real-time bed management, or coordinating care across a 20-hospital system needs architecture designed for scale.

Implementation Recommendations by Organization Type

Small practices usually start on Power Automate or Keragon, mid-size systems on UiPath or Power Automate, and large systems run a hybrid; startups prototype on self-hosted tools and build custom for production. Before picking any of them, choose one measurable workflow with our healthcare workflow automation guide.

Small Practices and Clinics (1 to 50 Providers)

Recommended: Microsoft Power Automate (if already on Microsoft 365) or Keragon (healthcare-first, BAA on every paid plan). Magical suits specific RCM tasks only after it signs a BAA. Small practices rarely have the DevOps capacity for self-hosted solutions and need turnkey HIPAA compliance. Power Automate's per-user pricing at $15/month is accessible, and the BAA is included by default.

Mid-Size Health Systems (50 to 500 Providers)

Recommended: UiPath or Power Automate for workflow automation; n8n self-hosted for flexible AI agent workflows. Mid-size organizations typically have IT teams capable of managing self-hosted infrastructure and benefit from the cost savings and flexibility of open-source platforms. UiPath's healthcare vertical provides pre-built templates that accelerate implementation.

Large Health Systems and IDNs (500+ Providers)

Recommended: A hybrid approach. Use Power Automate or UiPath for operational workflow automation. Deploy Flowise or n8n self-hosted on Azure/AWS for AI agent capabilities. Invest in custom development for clinical AI agents that require deep EHR integration, clinical reasoning, and regulatory compliance. Large systems have the engineering capacity to manage multiple platforms and the scale to justify custom development ROI.

Health Tech Startups and Digital Health Companies

Recommended: Flowise self-hosted for AI agent prototyping, n8n self-hosted for workflow automation, custom development for production. Startups need speed and flexibility but also need to build compliance into the product from day one. Self-hosted open-source platforms provide the best balance of rapid development, cost efficiency, and compliance readiness.

Related Reading

For the legal background, read HIPAA Was Not Built for AI: Legal & Technical Gaps.

• Building HIPAA-compliant AI agents: the architecture guide
• RPA vs agentic AI in healthcare: when to use which
• Build vs buy AI agents in healthcare
• Agent audit trails for HIPAA