Healthcare

Put clinical AI agents to work on ePHI and orders. Proof, a clinician's consent, and an audit trail on each.

Every agent gets a cryptographic identity. Every order or record it touches gets a clinician's consent at the moment of action, written to a tamper-evident record. The same proof covers your clinicians and help desk, on the Okta, Entra, or Ping you run.

One stolen password froze the nation's medical claims for months.
We make stolen credentials worthless.

This is the mechanism, not anyone's screen. A stolen credential hits a portal with the second factor off and walks into everything behind it. Origin-bound proof is welded to one device and one origin, so a stolen copy has nowhere to go.

WITHOUT ORIGIN-BINDINGWITH ORIGIN-BINDINGSECOND FACTOR OFFORIGIN-BOUND RAILstolen credentialclaims enginee-prescribingintegration busstolen credentialNOTHING TO REPLAYWelded to one device and origin.

At the doorWeb

Origin-bound proof. A stolen password and an attacker-in-the-middle proxy have nothing to replay.

Behind itMachine

Per-call signed identity on the integration accounts. No standing secret sits on the server to steal.

[1]

The breach this reconstructs (one portal, no MFA, a third of the country's claims down for months) is cited in the evidence rail, not drawn here.

The consequential action

An authenticated agent can still place the wrong order.
So a clinician signs, and the record is sealed.

A valid credential gets an agent in the door. It says nothing about whether the order is safe to place. So every consequential action takes a clinician's consent, bound to that order, and lands in a tamper-evident record. The EPCS-grade signing your prescribers already use.

Clinician consent gate and tamper-evident recordA clinical AI agent drafts a consequential order. The order meets a consent gate bound to its exact intent: a clinician signs at the moment of action, EPCS-grade and human in the loop. With consent the order proceeds and a signed entry is written to a tamper-evident, auditor-verifiable record. An order with no clinician consent is held at the gate and is not placed.clinical agentDRAFTED ORDER · INTENT-BOUNDCONSENT GATECLINICIAN SIGNSEPCS-GRADEORDER PLACEDsigned to recordAuditor-verifiable.TAMPER-EVIDENT RECORDHELDNo clinician consent. The order is not placed.

Actions·The consent

A clinician approves the action at the moment of signing, bound to that exact order. EPCS-grade, a human in the loop.

Actions·The record

Every order and approval is signed into a tamper-evident record an auditor can verify, without taking our word.

What's coming

The deepfake call is becoming the way in.
We answer with proof a clone can't fake.

A cloned-voice call to the help desk is already a documented healthcare attack. Gartner projects deepfakes will drive 40 percent of all social engineering by 2028. The knowledge questions and voiceprints that gate access today were built for a world without them.

Deepfake-driven social engineering, projected to 2028

Share of social engineering attacks using deepfake audio or video.

A single horizontal field represents all social engineering attacks. A held-red fill marks the 40 percent projected to use deepfake audio or video by 2028, with a dashed forecast boundary at the 40 percent line. The remaining 60 percent is a neutral share. Source: Gartner, September 2025.40%OF ALL SOCIAL ENGINEERINGDEEPFAKE-DRIVEN BY 2028DEEPFAKEREST OF SOCIAL ENGINEERING0%PROJECTED 2028100%

[2]Already documented in healthcare today: a cloned-voice call to the IT help desk, used to reset credentials and divert provider payments.

SOURCE . GARTNER, SEP 2025 (FORECAST)[3]

The kill chain

Walk the attacker's path.
We close every step.

The red-team view, not a scare story. Three steps, the way the real ones run, and the surface that closes each.

  1. 01

    A stolen credential meets a portal with no phishing-resistant MFA.[1]

    Intercept
    Web

    Origin-bound FIDO2/WebAuthn. The proof is welded to the real device and origin, so a stolen password has nothing to replay.

  2. 02

    A call to the IT help desk. A cloned voice. A password reset, and a redirected provider payment.[2]

    Intercept
    People

    Cryptographic person-to-person proofing, not knowledge questions an attacker already bought. The caller proves who they are, or they don't get in.

  3. 03

    A long-lived credential on an HL7/FHIR engine or a revenue-cycle bot, stolen and replayed.[4]

    Intercept
    Machine/Bot/Agent

    Per-call signed identity (RFC 7523). The private key never leaves the caller and the assertion expires fast, so there's no standing secret to steal.

The next credential that freezes the sector won't belong to a person.

The clinician test

A clinician will bypass anything that adds a step.
So we removed one.

A CMIO will block anything that adds a tap between a nurse and a patient, and should. So we didn't add one. We removed the password instead.

07:14 . shared nursing-station workstation

One tap, not a shared password on a sticky note.

One tap authenticates the clinician at a workstation a dozen people share, aligned to NIST SP 800-63-4 AAL3. Nothing to type, nothing to clone.

07:41 . controlled-substance order

Sign it, phishing-resistant.

The signature is bound to the prescriber, two-factor at the moment of signing, designed for EPCS-grade workflows. No token to fish out of a coat pocket.[5]

Designed for the EPCS-grade signing workflow your prescribers already use.

Coverage

Every surface a clinician, a caller, or a service touches.
One rail under all of them.

Web and Frontline carry the healthcare story, so they lead. The other six aren't a feature list. They're the rest of the attack surface, covered.

Web

Remote access and patient portals. Origin-bound.

Frontline

Shared clinical workstations. One tap, no shared password.

ActionsControlled-substance and high-risk order signing.
People

Help-desk proofing. A cloned voice can't reset a credential.

Bot

Revenue-cycle automation. Ephemeral tokens, full attribution.

Machine

HL7/FHIR engines. Every call signed.

Agent

Clinical and scheduling agents. Scoped and revocable.

Workload

Containerized workloads. Bound to where they run.

Voice

Patient verification without the data an attacker already has.

The regulatory picture

Regulators are mandating proof, not passwords.
We were built for it.

HIPAA's proposed rule makes MFA mandatory for ePHI. DEA already requires phishing-resistant signing for controlled substances. We cover the hard ones, and what comes next.

PROPOSED

HIPAA . Security Rule NPRM

HHS has proposed making MFA required, not addressable, for ePHI systems, alongside mandatory encryption. Still proposed, not final.

Phishing-resistant auth across every access point, aligned to NIST SP 800-63-4 AAL3. The encryption half is a separate control we sit alongside.

IN FORCE

DEA . EPCS

Two-factor at the moment a prescriber signs a controlled-substance order, with an auditable record. A signing event, not a session login.

Phishing-resistant signing built for EPCS-grade workflows, with the auditable record the rule wants.

IN FORCE

42 CFR . Part 2

Governs access to substance-use-disorder records, and now carries breach-notification weight.

Cryptographic proof of who authenticated, on which device, against which verifier. The trail an audit asks for.

Why now

The breaches and the rules are escalating together.

Every breach here pulled the next rule tighter. Two years of it, and what we cover against each.

  1. May 2024Incident

    Stolen credentials hit a remote-access portal with no MFA, then ransomware took a third of U.S. medical claims down for months.

  2. 2024Incident

    Attackers called hospital IT help desks with AI-cloned voices to reset credentials and divert provider payments.

  3. Published Jan 6, 2025Regulation

    HHS proposed making MFA mandatory, not optional, for every system that touches ePHI.

  4. 21 CFR 1311 Subpart CRegulation

    DEA requires two-factor at the moment a prescriber signs a controlled-substance order, with an auditable record.

Additional sources

  • GartnerSep 2025Estimate

    Gartner projects that by 2028, 40 percent of social engineering attacks will use deepfake audio and video, targeting executives and the broader workforce on voice and video calls.

    Gartner
  • The Hacker News (citing Rubrik Zero Labs 45:1 and Entro Labs 144:1)2025-2026Estimate

    Non-human identities (service accounts, API keys, tokens, workloads) outnumber humans by a wide margin, with reported ratios from about 45:1 to 144:1.

    The Hacker News (citing Rubrik Zero Labs 45:1 and Entro Labs 144:1)
  • HHS Office for Civil Rights breach portal2024

    HHS Office for Civil Rights logs every healthcare breach of 500 or more records. 2024 set a record for individuals affected, driven largely by a single clearinghouse breach.

    HHS Office for Civil Rights breach portal

How we fit

We make every identity in your stack provable.

A skeptical CISO has heard the pitch. Here is exactly how we slot in.

  • 01

    We prove who's at the keyboard. Your ePHI encryption protects the data at rest, and we sit alongside it.

  • 02

    We sign controlled-substance orders to EPCS-grade workflows, two-factor at the moment of signing.

  • 03

    Some legacy systems need a gateway in front to take a modern authenticator. We map that with you up front.

  • 04

    We federate into your Okta, Entra, or Ping and make every identity on them cryptographic.

Next step

Find the open door before an attacker does.
Book a 30-minute technical review.

Bring your remote-access topology, your help-desk reset flow, and your EHR and integration inventory. We'll show you exactly what we cover, and where you still need it.