BetaFeatures and data are not guaranteed while we are in beta. Do not rely on it as your system of record yet. Beta terms

Trust

Compliance posture

Last updated September 24, 2026

Kairos is a healthcare platform, and every part of it is designed against the frameworks a hospital's privacy officer, security team and counsel will ask about. Each control carries cross-framework keys, so one implementation produces evidence for all of them.

Honest status. Kairos is designed and evidenced against these frameworks; it is not yet certified. Live control status is published inside the product and summarised on the Compliance page.

How to read the status

Statuses use the legend of our compliance register.

Designed & evidenced
Implemented in the product; where a runtime check exists, it produces evidence continuously.
Infrastructure
Implemented by the cloud infrastructure stack: encryption, backups, TLS, image scanning.
At pilot
Required before real patient data enters the system. Tracked, and not needed for the synthetic demonstration.

Frameworks

  • HIPAA and HITECH

    Privacy, Security and Breach Notification Rules · implemented per NIST SP 800-66 r2

    Every Security Rule safeguard is mapped to where it lives: role-based access, a 15-minute automatic logoff, an encrypted identifier vault and an append-only audit trail.

    • Control ids cite the NIST SP 800-66 activity each satisfies.
    • Direct identifiers live only in an encrypted vault; one service may decrypt them, and every decrypt is logged.
    • Breach workflow follows the 60-day HHS notice and the 500-plus media notice.
    Designed & evidenced
  • HITRUST CSF v11

    e1 → i1 → r2

    Controls carry HITRUST domain tags so one implementation produces evidence for the assessment hospitals ask vendors for. Target: e1 readiness at pilot, i1 in year one.

    • Tagged domains include access control, transmission protection, password management, and data protection and privacy.
    • r2 is planned alongside the first large health-system deployment.
    At pilot
  • SOC 2 Type II

    Security · Availability · Confidentiality · Privacy

    Controls map to the Trust Services Criteria and evidence is collected automatically where possible. No audit has been performed yet; the first window opens after the pilot.

    • Change management runs through reviewed pull requests, CI gates and versioned database migrations.
    • Boundary protection: security headers, an allow-listed CORS policy and rate limiting.
    Designed & evidenced
  • ISO/IEC 27001, 27002 and 27799

    ISMS · Annex A · health informatics

    Annex A tagging yields a Statement of Applicability, and the health-informatics additions — patient-record audit trails, pseudonymisation, clinical-system availability — are built in.

    • Data-leakage prevention: a log scrubber and no identifiers in URLs.
    • Supplier relationships are tracked in a business-associate registry with agreement status per vendor.
    Designed & evidenced
  • ISO/IEC 42001 and NIST AI RMF

    AI management system · AI risk

    An inventory of every AI use with its intended purpose, a human decision on every recommendation, stored tool traces, and a bias review of risk adjustment.

    • Nothing the AI produces is applied automatically.
    • Every answer stores the tool calls that produced it, so a reviewer can see exactly which queries ran.
    Designed & evidenced
  • NIST CSF 2.0 and HHS 405(d)

    Govern · Identify · Protect · Detect · Respond · Recover

    The executive view of the register, with the health-sector cybersecurity practices for vulnerability management, backups, access and incident response.

    • Dependency scanning in CI and image scanning in the container registry.
    • Encrypted automated database snapshots with cross-region copies and point-in-time recovery.
    Infrastructure
  • California CMIA, H&SC §1280.15 and CCPA/CPRA

    Civil Code §56 · Health & Safety Code §1280.15 · Civil Code §1798.82

    The pilot state's stricter medical-privacy law, the 15-business-day clock for reporting unlawful access to the state health department, and consumer privacy for non-medical personal data.

    • The incident-response runbook has a California branch with the 15-business-day clock.
    • No marketing use of medical information, ever.
    At pilot
  • 21st Century Cures Act

    ONC information blocking

    Kairos never blocks export: every dataset is downloadable as a FHIR Bundle, NDJSON or CSV.

    • Kairos is not an EHR and is not ONC-certified, and it still treats export as a right, not a feature.
    Designed & evidenced
  • FDA Clinical Decision Support guidance

    Non-device CDS (2022)

    Kairos stays non-device software: it shows the basis for every recommendation, supports rather than replaces clinician judgment, and never analyses medical images or physiological signals.

    • Copy rules enforce this: "flagged for review", "recommended for release", never a clinical conclusion about a patient.
    Designed & evidenced
  • Anti-Kickback Statute, Stark Law and CMP gainsharing

    OIG gainsharing safeguards

    The safeguards are enforced in software: a safety gate, indicated-only stewardship, caps, independent validation, a documented method and attorney review before any program activates.

    • The value program is the platform's highest legal-risk feature, so it is designed to fail closed.
    Designed & evidenced
  • WCAG 2.1 AA and Section 508

    Accessibility

    Keyboard navigation, 4.5:1 text contrast, visible focus and labelled controls on every page — a hospital procurement requirement.

    • Charts carry text alternatives and data tables; status is never conveyed by colour alone.
    Designed & evidenced

Not applicable by design

PCI DSS — Kairos never touches payment card data. 42 CFR Part 2 — substance-use-disorder program records are excluded from the episode model.

Selected controls

A selection from the control catalogue that is seeded into the product. Customers see every control with its latest evidence inside Kairos.

Selected controls and their status
ControlImplementationStatus
Access controlHIPAA-164.312(a)(1)Roles, facility scoping and an explicit PHI grantDesigned & evidenced
Automatic logoffHIPAA-164.312(a)(2)(iii)15-minute idle timeout with a server-side session registryDesigned & evidenced
Encryption and decryptionHIPAA-164.312(a)(2)(iv)Identifier vault, AES-256-GCM, versioned keysDesigned & evidenced
Audit controlsHIPAA-164.312(b)Append-only, hash-chained audit and PHI access logDesigned & evidenced
Transmission securityHIPAA-164.312(e)(1)TLS-only load balancer with HSTSInfrastructure
Contingency planHIPAA-164.308(a)(7)Encrypted snapshots, cross-region copy, point-in-time recoveryInfrastructure
Business associate contractsHIPAA-164.308(b)Registry with agreement status per vendorDesigned & evidenced
De-identificationHIPAA-164.514(b)De-identified context and an outbound guard for every AI callDesigned & evidenced
Vulnerability managementSOC2-CC7.1Dependency scanning in CI, image scanning in the registryInfrastructure
Human oversight of AIISO42001-A.9Nothing auto-applied; recommendations need a personDesigned & evidenced
OIG gainsharing safeguardsAKS_STARK-gainsharingSafety gate, indicated-only stewardship, caps, validation, legal-review flagDesigned & evidenced
Privacy programHITRUST-19.aPrivacy officer role and approved policiesAt pilot
California authorisation rulesCMIA-56.10Privacy policy; no marketing use of medical informationAt pilot

AI governance commitments

Aligned with ISO/IEC 42001 and the NIST AI Risk Management Framework, and with HIPAA de-identification.

  • De-identified context only

    The AI layer can only receive de-identified aggregates. A structural type and a guard on every outbound request block identifiers, and a blocked call is logged.

  • Tool traces you can read

    Every answer stores the queries that produced it, one click from the answer.

  • Estimates you check

    AI-generated numbers are labelled as estimates, shown in a distinct colour, and carry their basis.

  • Nothing auto-applied

    Recommendations wait for a person. Releasing a block, booking a case or validating a saving is always a human action.

  • Data, not instructions

    Free-text fields never enter system prompts; they are passed as tool results explicitly framed as data.

Gainsharing safeguards

The value program is the platform's highest legal-risk feature because it connects hospital savings to clinician pay. The Anti-Kickback Statute, the Stark Law and the Civil Monetary Penalties Law govern it, so the OIG safeguards are enforced in software.

  • Safety gate

    No value payment in a period where a minimum safety or compliance threshold is missed.

  • Risk adjustment

    Outcomes are adjusted for procedure, acuity (ASA) and patient factors outside the team's control.

  • No single-metric payment

    Payment rests on a balanced scorecard; weights are configurable per hospital.

  • Indicated-only stewardship

    Stewardship counts only tests and consults the protocol says were not indicated, so avoiding necessary care is never rewarded.

  • No payment without validation

    Every savings line passes clinical review and finance review before it is shared.

  • Caps and a documented method

    Payouts are capped, the method is written down, and every program carries an attorney-review flag that must clear before activation.

Governance cadence

Once a pilot is live, the register is reviewed on a fixed rhythm.

Monthly
Review failed evidence checks and break-glass use; recertify access for PHI-granted users.
Quarterly
Update the risk register; review vendors and business associate agreements; test a restore from backup.
Annually
HIPAA risk analysis, penetration test, policy re-approval, workforce training renewal, HITRUST and SOC 2 assessment window.

The live register is inside the product

Customers' administrators see every control with its implementation, cross-framework keys and the timestamp of its latest evidence check, run by the platform itself. Security questionnaires and the named sub-processor list are available on request — see the security page.