Kendrake AI Assurance

Know which authorization evidence a change invalidates, before you make it.

AI systems change all the time. Kendrake AI Assurance keeps the model of your system, the requirements that apply to it and the evidence each part owes. When something changes, its Kendrake Delta engine tells you which evidence still holds, which must be redone, and why.

Kendrake AI Assurance change detail for change CHG-0006 on the fictional HARBORLIGHT system. It lists six attribute changes, the ten components they reached, and a panel titled "Why the engine decided what it did" with one review-required rule and several revalidate rules, each naming what triggered it.
HARBORLIGHT is a fictional AI system. Change detail. Every disposition names the rule that decided it. View full size: Change detail. Every disposition names the rule that decided it

Every change raises the same question.

Retrained models, new prompts, swapped providers, new data sources, agents given new permissions. Each one leaves the authorization team asking which of its hundreds of pieces of evidence are still valid.

Today that answer is a spreadsheet, a meeting and a guess. So teams either redo everything, which is slow and expensive, or redo too little, which puts the authorization at risk.

What Kendrake Delta does with a change

Delta is the change-impact engine inside Kendrake AI Assurance. It works the same way on every change.

  1. The change

    A retrained model, a new prompt, a swapped provider, a new data source or an agent with new permissions. Delta compares the new version of the system model with the last one.

  2. How far it reaches

    Delta shows which components the change reaches, directly or through the components that depend on them.

  3. The disposition and its rule

    Every piece of evidence gets a disposition: carry forward, revalidate, or send to a person to review. Each one names the rule that decided it. Same inputs, same answer, every time.

  4. The human decision

    Assessors and owners decide. Every decision, signature and change goes into a tamper-evident audit trail.

  5. The package

    Findings, risks and the POA&M flow out as OSCAL (SSP, assessment results, POA&M) and an eMASS-ready POA&M file.

Architecture map of the fictional HARBORLIGHT system with the impact of change CHG-0006 overlaid. Changed and affected components, including FailurePredictor v3, ManualAssistant pipeline, Read technical manuals and WorkOrderService, are outlined and marked revalidate or review required.
HARBORLIGHT is a fictional AI system. Architecture map with the impact of one change overlaid. View full size: Architecture map with the impact of one change overlaid

What you can check for yourself

Each of these can be verified in the product during a walkthrough.

  • Built on NIST content

    The requirements catalog and baselines are generated from NIST’s own published OSCAL content.

  • Valid OSCAL

    OSCAL output is validated against NIST’s OSCAL 1.1.2 schemas.

  • Deterministic and replayable

    The same change under the same rule pack gives a byte-identical result. Delta is checked against 40 change scenarios, 20 of them written by an independent author who never saw the rules.

  • Signed rule packs

    Rule packs are versioned and cryptographically signed. Requirements are data, not code.

  • Built for regulated hosting

    Runs in containers on Azure, Kubernetes or a customer’s own environment, with tenant isolation enforced in the database, encrypted evidence storage and a hash-chained audit log.

  • Separation of duties

    The person who uploads evidence cannot judge it. The system owner cannot sign their own package.

  • Accessible

    Tested against WCAG 2.1 AA in light and dark themes.

  • Tested and reviewed

    More than 500 automated tests, and an independent adversarial security review with every high finding fixed.

Government services

We deliver program, AI governance and authorization support to federal, state and local agencies, and to primes that need a small business teammate.

  • Program and product delivery

    Plans, schedules, risks and reporting for software and AI programs, run to a steady cadence with clear owners.

  • AI governance and RMF/ATO readiness support

    Map AI systems to NIST SP 800-53 and the NIST AI RMF, organize the evidence and prepare packages for assessors.

  • Agile delivery leadership

    Release planning, delivery governance and coaching for teams that build and ship software.

  • Technical evaluation support

    Structured review of vendor-delivered AI work against requirements, written so the government can act on it.

Company facts

Company
Kendrake LLC
Maryland limited liability company, small business
UEI
SJ6KBLMNXAQ3
NAICS
541611 (primary), 541512, 541618, 541690, 541990, 611430
Location
Gaithersburg, Maryland
Contact
[email protected]
kendrake.com

Defense subcontractor? Start with the free tools.

Your primes now write AI rules into their supplier terms. These tools show where you stand, and the AI-Use Readiness Sprint builds the record your prime will ask for.

See Kendrake Delta on a fictional system.

A 20-minute walkthrough on HARBORLIGHT, a fictional AI system: a model retrain previewed before it is saved, what Delta reopens and why, and the package an assessor signs.