HIPAA compliance and AI governance, run as one program
A healthcare organization answers two sets of questions about the same systems. HIPAA asks where its protected health information is and whether it is safeguarded; AI governance asks which AI systems it runs, how much scrutiny each needs, and who approved each one and why. This page explains what each program involves and why the two work better as one.
Free plan, no card needed, unlimited team members. Watney, exports and scheduled report emails, auditor and guest links, the transparency page and public badge, the API and MCP server, webhooks and SIEM forwarding, and SAML single sign-on and SCIM provisioning come with a paid plan; the download of all your organization's data works on every plan.
The parts of the program
In a healthcare organization the same systems carry two sets of obligations. The HIPAA program finds where ePHI is held, judges what threatens it and keeps the records the rules call for. The AI governance program finds where AI is used, whether a team built it, a vendor sells it or it arrived as a feature inside software you already run, and records who approved each use and why.
One register of systems
You cannot govern or protect a system nobody listed. One register records each system once, whether it handles ePHI, whether it uses AI, the devices and models behind it, and the vendors that supply it, with your due diligence on each vendor.
A HIPAA risk analysis
The Security Rule asks for an accurate and thorough assessment of the risks to the ePHI you hold, kept current as your systems change, the AI tools that handle ePHI included.
Assessment against a framework
A framework gives the program one consistent set of questions: the HIPAA rules for a HIPAA assessment, and a framework such as the NIST AI RMF for AI governance, asked of the organization as a whole and of each system.
A risk tier for each AI system
Not every system needs the same scrutiny. A risk tier sets how closely a system is reviewed and how long an approval lasts before it is revisited.
Breach notification
An incident that involved PHI gets a breach risk assessment. A breach of unsecured PHI lists the notices it owes, each with the deadline the rule sets and the date it was given: for a covered entity, to individuals, HHS and sometimes the media; for a business associate, to the covered entity.
Review and approval
Before a system that uses AI goes into use, named people decide whether it should, and each records a reason. Each approval carries a review date set by its risk tier; when it arrives a re-review opens automatically, so the decision is made again as the system and its context change.
Requests, disclosures and business associates
Requests from individuals, with the deadline the rule sets where it sets one; a log of disclosures to account from; and each vendor that creates, receives, maintains or transmits PHI for you, with the state of its business associate agreement.
Model and system cards
Cards describe a model or a system in a form a reviewer, an auditor or the public can read. A published version is fixed so it can be relied on later, and publishing system cards lets the people a system affects see what it is and how it is governed.
Records that stay current
Risks, incidents, policies, training, changes and access reviews, kept next to the systems they concern, with an audit trail of who changed what.
Who makes each decision
Watney, the AI assistant, drafts cards, suggests triage and answers questions. The approval, the risk tier and the decision to publish belong to the people your policy names.
What a program does and does not show. HIPAA has no certificate, and running a program does not by itself mean an organization meets any law or standard. Backsplice supports the program: it keeps the register of systems, runs the assessments, routes the reviews and records the decisions. The judgments stay with your organization and its advisers.
Both programs start with a list of what you run
Every other part of the program depends on the register. A system missing from it goes unassessed and unreviewed, nobody can say who approved it, and nothing prompts you to include its ePHI in the risk analysis. The NIST AI RMF puts the inventory among its first Govern outcomes:
“Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities.”NIST AI RMF 1.0, GOVERN 1.6
Quoted from NIST's published framework, which is voluntary guidance.
Each entry records more than a name:
- Whether it handles ePHI, and whether it uses AI. One system, two marks: the HIPAA questions follow the one, the AI governance questions the other, and a system with both carries both.
- How the system is sourced. Built in-house, vendor AI product, AI feature inside other software. The last is the easiest to miss.
- The models it uses, with versions. So a model change shows in the record.
- The vendors behind it. Including whether a vendor trains on your data and whether an agreement is on file.
The health information your systems hold
For a covered entity or a business associate, HIPAA's obligations follow protected health information into every system that creates, receives, maintains or transmits it, AI tools among them. The program keeps the records the HIPAA rules call for next to the systems they concern.
The risk analysis
An assessment of the risks to the ePHI you hold, and the measures that reduce them, kept current as your systems change.
Business associates
An agreement with each vendor that handles PHI for you, with its state kept on the vendor's record.
Breach notification
A risk assessment when PHI may have been compromised, and notice to individuals, HHS and sometimes the media by the deadlines the rule sets.
Individuals' requests
Access, amendment and an accounting of disclosures, each answered within the time the rule allows, and requests for restrictions and confidential communications.
Read the guide to HIPAA for who it covers and what each rule asks. Backsplice signs a business associate agreement with an organization that says it is a covered entity or a business associate.
More scrutiny where more can go wrong
A low-risk tool and a system that makes decisions about people need different scrutiny. The NIST AI RMF asks organizations to set the level of risk management activity by their own risk tolerance (GOVERN 1.3). A risk tier is how a program does that.
More approvers, a shorter approval
A system that makes automated decisions about people, or that customers use directly, may call for more approvers, agreement from all of them, and a sooner re-review.
Fewer approvers, a longer approval
An internal assistant whose output a person checks may need a single approver and a longer approval period. It is still registered, assessed and recorded like every other system.
In Backsplice, a system's risk tier sets who must approve it and whether any, a majority or all of them must agree, under your risk-tier policy (a weaker per-system policy needs an org admin and a reason). An approval lasts for a period set by the tier (High: 6 months; Moderate: 12 months; Low: 12 months; Unclassified: 6 months), then prompts a re-review.
A framework gives the program its questions
A framework gives every review the same questions. A HIPAA assessment asks questions drawn from the Security, Breach Notification and Privacy Rules, which are law, of your program and, where they apply, of each system that handles ePHI. On the Free plan a HIPAA assessment asks the HIPAA Security and Breach Notification Rules questions only. The NIST AI Risk Management Framework is voluntary guidance from the US National Institute of Standards and Technology, organized into four functions: Govern, Map, Measure and Manage. NIST AI 600-1, the Generative AI Profile, adds the risks particular to generative AI. Security frameworks for AI, such as the OWASP Top 10 for LLM Applications and the mitigations in MITRE ATLAS, ask how each system is protected against attack and misuse. Each of these is voluntary guidance.
A program answers some questions once, for the organization, and others for each system, and each assessment is of one kind, HIPAA or AI governance, scored on its own. In Backsplice the program assessment asks the governance questions, each system assessment asks the questions about that system, and the generative AI questions are asked of each system that uses generative AI, and in the program assessment while any live system does: the NIST AI 600-1 (Generative AI Profile) questions, and the Singapore Model AI Governance Framework for Generative AI (2024), Singapore Model AI Governance Framework for Agentic AI (v1.5, 2026) and OWASP Top 10 for LLM Applications 2026 questions where an assessment selects those packs.
Backsplice includes NIST AI RMF 1.0, NIST AI 600-1 (Generative AI Profile), NIST Cyber AI Profile (IR 8596 initial preliminary draft), NIST COSAiS Predictive AI (built from NIST's January 2026 outline), ISO/IEC 42001:2023, ISO/IEC 42005:2025, ISO/IEC 23894:2023, EU AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), EU AI Act GPAI model obligations (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), NYC Local Law 144 of 2021 (AEDT), Colorado ADMT Act (C.R.S. 6-1-1701 to 6-1-1709, SB 26-189), California CCPA Regulations on ADMT and Risk Assessments (2025), California Civil Rights Council Regulations on Automated-Decision Systems (2025), Illinois Human Rights Act (775 ILCS 5), as amended by P.A. 103-0804 (HB 3773), Texas Responsible Artificial Intelligence Governance Act (H.B. 149, 2025), Utah AI Laws (Utah Code 13-72, 13-72a, 13-77 and 58-60-118, as amended through 2026), UK AI Cyber Security Code of Practice (2025), Singapore Model AI Governance Framework (2nd ed., 2020), Singapore Model AI Governance Framework for Generative AI (2024), Singapore Model AI Governance Framework for Agentic AI (v1.5, 2026), OWASP Top 10 for LLM Applications 2026, MITRE ATLAS 2026.09, HIPAA Security and Breach Notification Rules (45 CFR 164 Subparts C–D, eCFR 2026-10-06), HIPAA Privacy Rule (45 CFR 164 Subpart E, eCFR 2026-10-06), 42 CFR Part 2 (Substance Use Disorder Patient Records, eCFR 2026-10-05), HHS 405(d) HICP (2023 Edition), NIST SP 800-66 Rev. 2 (February 2024) and ONC Information Blocking (45 CFR 171, eCFR 2026-10-06) on Premium and Enterprise. The Free plan includes them too, except the HIPAA Privacy Rule and the optional HIPAA packs: a HIPAA assessment on it asks the HIPAA Security and Breach Notification Rules questions only.
13 packs are drawn from law (statutes and regulations); the others are voluntary guidance or standards. They are not legal advice: whether and from when a law applies to your organization is for it to settle, with counsel where needed.
Every decision has an approver and a reason
A review ends in a decision with an owner. The NIST AI RMF describes it as a determination of whether a system should go ahead:
“A determination is made as to whether the AI system achieves its intended purposes and stated objectives and whether its development or deployment should proceed.”NIST AI RMF 1.0, MANAGE 1.1
Every decision carries a reason, each approval opens a re-review when its review date arrives, and the approval dossier exports the whole record when someone asks how a system was approved.
Write down what each system is, and let people read it
An approval records a decision. A card records what was decided about: what the model or system is, what it is for, and what its owners know about it, written down where a later reader can find it.
- Model cards and system cards. One for each model you record and each system you run.
- A fixed version once published. In Backsplice, publishing freezes that version with a tamper-evident seal, so a later edit makes a new version.
- A public transparency page. On a paid plan, list your published system cards where the people they affect can read them.
Priced by the systems you register, with no per-seat fees
The people who own a system, the people who review it, and the people who approve it all need to be in the record. So we do not charge per seat: every plan, the Free plan included, has unlimited team members.
Premium includes 1 AI system and 1 system that handles ePHI. Enterprise is $499 a month or $4,999 a year with no limit on either. The features are the same on both paid plans, and Watney's fair-use limit scales with the plan.
Unlimited team members
Everyone who owns, reviews or approves a system can be in the record, on every plan.
HIPAA and AI governance
Both kinds of assessment, and the records each calls for, in one program on every plan. The Free plan includes 1 AI system and 1 system that handles ePHI, and a HIPAA assessment on it asks the HIPAA Security and Breach Notification Rules questions only; the Privacy Rule questions and the optional HIPAA packs come with a paid plan, and answers already given are kept.
A business associate agreement
For an organization that says it is a covered entity or a business associate. Our standard agreement, which you can read on our site, is signed by Backsplice in advance and takes effect when you sign it; or you can provide your own, which we review and countersign.
The framework packs
NIST AI RMF 1.0, NIST AI 600-1 (Generative AI Profile), NIST Cyber AI Profile (IR 8596 initial preliminary draft), NIST COSAiS Predictive AI (built from NIST's January 2026 outline), ISO/IEC 42001:2023, ISO/IEC 42005:2025, ISO/IEC 23894:2023, EU AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), EU AI Act GPAI model obligations (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), NYC Local Law 144 of 2021 (AEDT), Colorado ADMT Act (C.R.S. 6-1-1701 to 6-1-1709, SB 26-189), California CCPA Regulations on ADMT and Risk Assessments (2025), California Civil Rights Council Regulations on Automated-Decision Systems (2025), Illinois Human Rights Act (775 ILCS 5), as amended by P.A. 103-0804 (HB 3773), Texas Responsible Artificial Intelligence Governance Act (H.B. 149, 2025), Utah AI Laws (Utah Code 13-72, 13-72a, 13-77 and 58-60-118, as amended through 2026), UK AI Cyber Security Code of Practice (2025), Singapore Model AI Governance Framework (2nd ed., 2020), Singapore Model AI Governance Framework for Generative AI (2024), Singapore Model AI Governance Framework for Agentic AI (v1.5, 2026), OWASP Top 10 for LLM Applications 2026, MITRE ATLAS 2026.09, HIPAA Security and Breach Notification Rules (45 CFR 164 Subparts C–D, eCFR 2026-10-06), HIPAA Privacy Rule (45 CFR 164 Subpart E, eCFR 2026-10-06), 42 CFR Part 2 (Substance Use Disorder Patient Records, eCFR 2026-10-05), HHS 405(d) HICP (2023 Edition), NIST SP 800-66 Rev. 2 (February 2024) and ONC Information Blocking (45 CFR 171, eCFR 2026-10-06), on Premium and Enterprise. The Free plan includes them too, except the HIPAA Privacy Rule and the optional HIPAA packs: a HIPAA assessment on it asks the HIPAA Security and Breach Notification Rules questions only.
Watney
Included on Premium and Enterprise at no extra charge, within fair-use limits. Not part of the Free plan.
SAML single sign-on and SCIM
On Premium and Enterprise. Not part of the Free plan.
A data processing agreement
An organization admin accepts it for the organization before any governance records can be created or changed. Organization setup (profile, members, sign-in and integration settings) can be done first.
Exports, API and transparency page
Watney, exports and scheduled report emails, auditor and guest links, the transparency page and public badge, the API and MCP server, webhooks and SIEM forwarding, and SAML single sign-on and SCIM provisioning come with a paid plan; the download of all your organization's data works on every plan.
See full pricing and plan comparison, or the product overview.
Questions about HIPAA compliance and AI governance
Why run HIPAA compliance and AI governance as one program? +
In a healthcare organization they concern the same systems and the same vendors. An AI tool that handles protected health information belongs in the HIPAA risk analysis and, where its vendor handles PHI for you, needs a business associate agreement with that vendor, and the same tool needs a risk tier, a review and an approval. Run separately, the two programs keep two lists of the same systems. Backsplice keeps one register of systems, marked as handling ePHI, using AI, or both, and scores each kind of assessment on its own.
What does HIPAA ask of an organization that adopts AI? +
The same as of any other system that handles protected health information. An AI tool that creates, receives, maintains or transmits ePHI belongs in the Security Rule risk analysis; a vendor whose AI service handles PHI for you is generally a business associate, which needs a business associate agreement; and the minimum necessary standard applies to what your workforce puts into it, unless one of its exceptions does. HIPAA does not regulate AI as such. Our guide to HIPAA cites the rule behind each of these. Nothing here is legal advice.
Who do the HIPAA rules apply to? +
Covered entities, which are health plans, health care clearinghouses, and health care providers that transmit health information electronically in connection with a transaction HIPAA covers, such as a claim; and their business associates, a person or company, other than a member of the covered entity's workforce, that creates, receives, maintains or transmits PHI on a covered entity's behalf, or provides it certain services that involve PHI, such as legal, accounting or consulting services. A subcontractor that creates, receives, maintains or transmits PHI for a business associate is a business associate too. A business associate is directly bound by the Security Rule and by the parts of the Privacy and Breach Notification Rules that name it, and by its business associate agreement. Our guide to HIPAA cites the rule behind each. Nothing here is legal advice.
What does the Security Rule risk analysis involve? +
The Security Rule requires an accurate and thorough assessment of the potential risks and vulnerabilities to the confidentiality, integrity and availability of the ePHI you hold. In practice it covers every system that handles ePHI, the AI tools among them, names the reasonably anticipated threats and vulnerabilities with the likelihood and impact of each, is written down, and is reviewed and updated as your systems and operations change. In Backsplice the analysis is built from threat scenarios and approved, and any scenario can be added to the risk register, where the work to reduce it is tracked. Approving it records your own review; whether it is the analysis the rule requires is for your organization to decide.
What does a breach of unsecured PHI require? +
Apart from three narrow exclusions, an acquisition, access, use or disclosure of PHI that the Privacy Rule does not permit is presumed to be a breach unless the covered entity or business associate shows, through a documented risk assessment of at least four factors, a low probability that the PHI has been compromised. Notice is owed when the PHI was unsecured, meaning not rendered unusable, unreadable or indecipherable under HHS guidance, such as by encryption. A covered entity then notifies the affected individuals without unreasonable delay and no later than 60 calendar days after discovery. It notifies HHS at the same time when 500 or more individuals are affected, and otherwise no later than 60 days after the end of the calendar year in which the breach was discovered. It notifies prominent media outlets when more than 500 residents of a State or jurisdiction are affected. A business associate notifies the covered entity without unreasonable delay and no later than 60 calendar days after discovery, and its agreement may set a shorter time. A law enforcement delay the rule allows can postpone these notices, and state breach laws can add their own requirements. In Backsplice an incident that involved PHI gets that risk assessment, and a breach lists each notice it owes with its deadline. Nothing here is legal advice.
What is AI governance? +
AI governance is the set of practices an organization uses to decide which AI systems it will use and how. In practice that means keeping an inventory of AI systems, judging how much risk each one carries, assessing them against a framework, having named people approve each system before it is used, documenting what each system is, and revisiting those decisions as things change.
Is AI governance required by law? +
That depends on where you operate, what your AI systems do, and your sector, and the law in this area is changing quickly. Many organizations run a program on a voluntary framework such as the NIST AI RMF whatever their legal position, because it gives them one consistent record. Backsplice does not decide which laws apply to you, and nothing here is legal advice.
What is the NIST AI RMF? +
The AI Risk Management Framework is voluntary guidance published by the US National Institute of Standards and Technology. It is organized into four functions: Govern, Map, Measure and Manage. NIST AI 600-1, the Generative AI Profile, is its companion for generative AI. Backsplice includes both among its framework packs: NIST AI RMF 1.0, NIST AI 600-1 (Generative AI Profile), NIST Cyber AI Profile (IR 8596 initial preliminary draft), NIST COSAiS Predictive AI (built from NIST's January 2026 outline), ISO/IEC 42001:2023, ISO/IEC 42005:2025, ISO/IEC 23894:2023, EU AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), EU AI Act GPAI model obligations (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), NYC Local Law 144 of 2021 (AEDT), Colorado ADMT Act (C.R.S. 6-1-1701 to 6-1-1709, SB 26-189), California CCPA Regulations on ADMT and Risk Assessments (2025), California Civil Rights Council Regulations on Automated-Decision Systems (2025), Illinois Human Rights Act (775 ILCS 5), as amended by P.A. 103-0804 (HB 3773), Texas Responsible Artificial Intelligence Governance Act (H.B. 149, 2025), Utah AI Laws (Utah Code 13-72, 13-72a, 13-77 and 58-60-118, as amended through 2026), UK AI Cyber Security Code of Practice (2025), Singapore Model AI Governance Framework (2nd ed., 2020), Singapore Model AI Governance Framework for Generative AI (2024), Singapore Model AI Governance Framework for Agentic AI (v1.5, 2026), OWASP Top 10 for LLM Applications 2026, MITRE ATLAS 2026.09, HIPAA Security and Breach Notification Rules (45 CFR 164 Subparts C–D, eCFR 2026-10-06), HIPAA Privacy Rule (45 CFR 164 Subpart E, eCFR 2026-10-06), 42 CFR Part 2 (Substance Use Disorder Patient Records, eCFR 2026-10-05), HHS 405(d) HICP (2023 Edition), NIST SP 800-66 Rev. 2 (February 2024) and ONC Information Blocking (45 CFR 171, eCFR 2026-10-06).
Why keep a register of systems? +
Every other part of the program depends on it. You cannot assess, approve or protect a system nobody listed. AI also arrives in more ways than a team building a model: a vendor product, or a feature switched on inside software you already use. A register records each system, whether it handles ePHI and whether it uses AI, the models it uses and their versions, and the vendors that supply them.
What is a risk tier for? +
It keeps scrutiny proportionate. A system that makes automated decisions about people may deserve a wider review than an internal drafting assistant a person checks. In Backsplice, the risk tier sets who must approve a system, whether any, a majority or all of them must agree, and how long the approval lasts (High: 6 months; Moderate: 12 months; Low: 12 months; Unclassified: 6 months).
What are model cards and system cards? +
Short documents that describe a model or an AI system in a form other people can read. In Backsplice, publishing a card freezes that version with a tamper-evident seal. A later edit creates a new version and leaves the published one as it was. Published system cards can be listed on a public transparency page on a paid plan.
Does the AI make any of the decisions? +
No. Watney, the AI assistant in Backsplice, drafts cards, suggests intake triage, and answers questions grounded on your records. It does not approve a system, set a risk tier, or publish anything. An organization admin must opt in before Watney reads your data, and it sends the model provider only what a task needs; researching a model card also runs web searches the model writes from your request and the card. Its own system card names the provider and the models.
What does it cost, and are there per-seat fees? +
No per-seat fees, with unlimited team members on every plan. Start on the Free plan, no card and no time limit. Premium is $99 a month or $999 a year and includes 1 AI system and 1 system that handles ePHI; Enterprise is $499 a month or $4,999 a year with no limit on either. The features are the same on both paid plans, and Watney's fair-use limit scales with the plan.
Not legal advice. Backsplice is software for running a HIPAA and AI governance program. We are not a law firm, and nothing on this page is legal advice. 13 packs are drawn from law (statutes and regulations); the others are voluntary guidance or standards. They are not legal advice: whether and from when a law applies to your organization is for it to settle, with counsel where needed. The NIST Cyber AI Profile is a NIST draft. Which laws apply to you depends on your organization, where it operates and what its systems do; have your counsel review what you rely on.