System · PIPAA — AI Compliance Copilot
Ask HIPAA questions. Without handing your PHI to a third-party model.
PIPAA runs on inference infrastructure Patient Protect controls, so prompts and PHI stay inside the platform rather than traveling to a consumer model API. Air-gapped hardware at your practice is a separate thing and it is in development. Architectural compliance, not asserted.
Basic: PIPAA at the standard usage level. Pro: PIPAA at a higher usage level.
HIPAA mapping
Where this fits in the HIPAA rules.
4 provisions this capability contributes to, each with the specific PIPAA — AI Compliance Copilot behavior behind it. The obligation stays with your practice — the mapping shows which part of the work the platform carries.
§164.502(a)Uses and disclosures of PHI
The question worth asking of any AI assistant is where the prompt goes. PIPAA's inference runs on infrastructure Patient Protect operates as your business associate under the agreement you already hold with us, rather than being passed to a consumer model API whose terms you have not negotiated and whose retention you cannot inspect.
§164.514(a)De-identification standard
De-identification is one route to putting information outside the Rule's reach, and it is the route practices are usually advised to take before pasting anything into a general-purpose assistant. Keeping inference inside a business-associate relationship is a different route to the same caution. Neither removes your own judgement about what is appropriate to ask.
§164.312(a)(1)Access control
Requires technical policies allowing access only to authorized persons. PIPAA sits inside the platform those controls govern rather than beside it, which is the point of running the assistant here instead of pasting practice detail into a general-purpose one. What any individual member sees in an answer is an authorization question worth confirming against your own account rather than inferring from this page.
§164.308(a)(4)Information access management
PIPAA's access patterns are logged in the same audit trail as every other PHI interaction.
What it does
The AI question, answered architecturally.
The question worth asking of any assistant is where the prompt goes, because for a general-purpose one the answer is a third-party model API. Your question (which usually contains the patient context that makes it worth asking) leaves your environment, gets processed by a third party, and may be retained for training. That is not HIPAA-compliant. It cannot be made HIPAA-compliant by policy.
PIPAA answers it with architecture rather than policy. Its inference request is handled on infrastructure Patient Protect controls, so the prompt is not sent to a third-party model API for inference, and the whole thing sits inside the agreement you already hold. That claim is deliberately the size it looks: it describes the inference request, not every path data takes through a system. PIPAA Air-Gapped, which would put the model on hardware at your office, is in development. Ask PIPAA, the free demo on this site, is an illustration rather than this product — it runs on a third-party cloud model and never sees a practice, which is why patient information stays out of it.
Ask the Copilot a compliance question in plain English. Two kinds of context sit behind the answer. The regulatory half is the rule itself, cited rather than paraphrased. The other half is your practice — inside Patient Protect, PIPAA can reason against the state the platform already holds for your office: the assessment and what it found, the policies you have adopted, your workforce, your systems inventory, your vendors and the agreements covering them, the advice queue and what has been recorded against it. It is the difference between asking what the rule says and asking what it means here. When the Copilot doesn't know, it says so. It does not hallucinate citations.
How it works
5 mechanisms keep PIPAA — AI Compliance Copilot working.
Two deployment modes — both keep PHI away from third-party LLMs.
PIPAA Cloud is generally available and runs on Patient Protect's secure inference infrastructure: prompts and practice data stay on infrastructure Patient Protect controls rather than being passed to a consumer model API. PIPAA Air-Gapped is in development (waitlist available): as designed, it would run on a Mac Mini M-series (or compatible) deployed at your office with zero outbound network and zero third-party calls. Both designs keep prompts and PHI away from third-party model providers; only PIPAA Cloud is available today.
It reasons over your practice, not a generic one.
Inside Patient Protect the Copilot works against the state the platform holds for your office — the assessment and what it found, your adopted policies, the workforce roster, the systems inventory, the vendors and their agreements, the advice queue and what has been recorded against it. That is what separates it from asking a general-purpose assistant a HIPAA question: the answer is about this practice. The edge of it is worth stating too — it knows what Patient Protect knows, so a vendor nobody entered or a system nobody inventoried is invisible to it, as it would be to anything.
CFR-grounded citations.
Every regulatory claim is cited with a § reference. The Copilot will not assert a HIPAA requirement without the citation. When the citation is contested or evolving, the Copilot says so.
It reasons; it does not act.
PIPAA reads your practice state and answers from it. It does not create work, change an assessment answer, edit a policy or close an item. Deciding what matters next is the Autonomous Compliance Engine's job and doing it is a person's, which is a distinction this category blurs constantly — an assistant that can quietly change the record is a different product with a different risk profile.
Used inside the agreement you already hold.
The in-platform Copilot runs under the business associate agreement between your practice and Patient Protect, on inference infrastructure we control. That is what makes reasoning over real practice data a governed act rather than a policy exception — and it is the substantive difference between this and the free public version on our website, which is connected to no account and must never be given patient information.
Who this is for
Built for the practices that need it most.
Practices that have ruled out cloud AI for compliance reasons.
If your compliance team or counsel has said no to ChatGPT, no to Claude.ai for clinical contexts, no to “we'll use the API with a BAA” — PIPAA is the answer that doesn't require trusting the provider, because the provider isn't in the loop.
Practices that want AI without the AI risk.
Cloud AI assistants are getting capable enough that workforce members will use them with or without office sanction. The shadow AI risk in healthcare is meaningful. PIPAA gives the workforce a sanctioned tool that's actually compliant — reducing the incentive for unsanctioned alternatives.
Specialty practices with high-sensitivity data.
Behavioral health, substance use, reproductive health, HIV care, and similar sectors operate under additional confidentiality restrictions (42 CFR Part 2 in some cases). The cloud-AI compliance gap is widest here. Keeping inference inside a governed business-associate relationship matters most in exactly these contexts — and it is worth adding that several of them carry confidentiality rules stricter than HIPAA, which no product setting resolves for you.
What you get
5 outcomes you’ll feel in week one.
Prompts stay inside the platform.
Inference runs on infrastructure Patient Protect controls, not a consumer model API.
Plain-English access to your compliance state.
Workforce members get answers without learning the platform's UI inside out.
CFR-cited responses.
Every regulatory claim is citation-backed. The Copilot will not bluff.
Sanctioned alternative to shadow AI.
Reduces the workforce pull toward non-compliant tools.
Architectural compliance, not asserted.
Compliance is verifiable in the deployment topology, not in a vendor BAA. Auditors see exactly where data goes — and where it doesn't.
What's the difference between PIPAA Cloud and PIPAA Air-Gapped?
What hardware does PIPAA Air-Gapped require?
How accurate is the model?
What if PIPAA doesn't know an answer?
Can PIPAA write policies for us?
Is PIPAA included in Basic?
Does the public “Ask PIPAA” demo on this site use the same architecture?
What does “PIPAA” stand for?
Continue exploring
Related features in the platform.
System
Autonomous Compliance Engine
Your practice state produces the next thing worth doing — from the assessment, but also from a lapsing BAA, a policy change, a new workforce member. Work closes when its conditions are met, and closing it changes the state.
Learn moreSystem
Risk Intelligence
Likelihood against impact, item by item, so a 331-question assessment comes back as a ranked list rather than a document.
Learn moreIntelligence
ePHI Data Flow Mapper
A free tool, in your browser, with no account. Place your systems and vendors, connect what talks to what, and the connections nobody has papered become the ones you are looking at.
Learn moreArchitecture does the compliance work for you.
The hardware sits on a shelf in your office. The model answers in seconds. The compliance is structural, not asserted.