VALIDANT AI
    Trustworthy AI, within everyone's reach

    Make AI fair, explainable, provable and accountable: make AI trustworthy.

    Expert-grade AI assurance, built for every team.

    Check a seal yourself

    We test AI systems for discrimination, explain their decisions in plain language, and turn the result into a proof anyone can check. For the governance you need today, and for a future where AI agents act on our behalf.

    Decentralized Identity FoundationTrust Over IP FoundationEcosystemGlinz & CompanyIceberg
    How it works

    Measure. Explain. Prove.

    The same three steps for a hiring model, a credit score or an AI agent. Each one ends in evidence someone else can check, not a slide someone has to believe.

    1. Measure

      We test whether an AI system treats groups of people differently, across 47 fairness metrics and statistical tests, with the evidence to show how sure we are.

      You getA plain-language verdict, with the numbers behind it.

    2. Explain

      We show which factors drive each decision and by how much, so “the model is unfair” becomes “this feature drives it, and by this much”.

      You getExplanations a regulator, a manager and the person affected can each read.

    3. Prove

      We seal the result as a signed credential. Anyone can check it is authentic, unaltered and not revoked, with no login and without trusting us. It covers one system at one moment, not every decision forever.

      You getA Trust Seal for the system, a Decision Receipt for the person.

    Illustrative data

    A model scores 40 applicants from each of two groups with the same spread of qualifications. Move the shortlist line and watch who gets through.

    Group A

    19 of 40 shortlisted

    Group B

    7 of 40 shortlisted

    Selection-rate ratio

    0.37

    Group A 48% · Group B 18%. Group B is shortlisted at 0.37× the rate.

    What drives the gap

    Postcode. It stands in for group membership and costs group B points it did not earn.

    A result that fails is not sealed.

    The four-fifths rule is one screening test, not the whole fairness question. A real audit runs 28 modules with confidence intervals, not one ratio.

    Three bodies · one assurance

    Every AI decision touches three parties. We assure all three.

    The model that decides, the organisation that answers for it, and the person it decides about. Trust holds only when all three can be checked, never one alone.

    Why trust is an orbit, not a pillar
    Metaphor

    The hall of due process

    Fairness in ML is not a feeling, it is a corridor of evidence. Each column is a fairness metric the model must pass through before it reaches a decision. The figure walking toward the vanishing point is the model under audit.

    the model
    01Live
    fair.AI fairness & explainability

    What your model actually does, with the receipts to prove it.

    Detect, mitigate, explain and monitor bias across the whole pipeline, from training data to LLMs. SHAP, LIME and counterfactuals turn “the model is unfair” into “this feature drives it, and by this much”, with the statistics to back it and EU AI Act mapping built in.

    • 28 assessment modules across the full lifecycle
    • 43 discrimination patterns · US / EU / CH jurisdictions
    • Explainable AI built in · SHAP, LIME and counterfactuals
    • Bootstrap, Bayesian and permutation testing
    See in the platform
    Metaphor

    The Swiss Federal Palace

    Civic institutions are where digital trust either holds or breaks. We chose Bern because Switzerland is where Validant is built and where digital sovereignty is taken as a serious craft, not a slogan.

    the organisation
    02Service today
    accountable.Digital trust

    Proof that your organisation keeps its promises.

    Built on the iceberg.digital framework, we audit whether your digital systems behave as promised: in governance, in engineering, and for the people who use them. Most of that work happens below the waterline, which is where trust is actually won or lost.

    • Ten trust constructs · 128 measurable cues
    • Governance, engineering, agency and institutional layers
    • Explicit and implicit signals, continuously assured

    We support digital-trust audits today, as a service. The platform modules are in progress.

    Metaphor

    The credential in your hand

    Identity should fit between your fingers, not in someone else's database. The hand is yours. The card is the proof. The city behind it is everywhere it can be presented and verified, with no platform in between.

    the person
    03Planned
    yours.Identity & agency

    A record of every AI decision, held by the person it is about.

    A Decision Receipt gives the person affected their own copy: why the decision fell as it did, and how to ask for a human review. Next come portable credentials and signed agent mandates, so an agent acting for you can prove what it may do, and you can take that back.

    • Decision Receipt in the person's own wallet
    • Trust Experience Library of first-person patterns
    • Credentials and signed agent mandates on the roadmap
    Films

    Where an algorithm already decides.

    Four short films, four places where AI already makes the call: what a child is shown, who gets the job, who gets the loan, who gets care. Pick a use case.

    Choose a use case

    Hiring · Full film · 2:41

    Trust me, I'm an algorithm. A self-driving car that has read the entire internet, mostly the comment sections, screens Maria's job application. Postcode. Rejected. You can outsource the decision. Not the responsibility.

    Share this film
    Open standards · VTI

    An open trust layer, for people and the AI that acts for them.

    Whether a person proves who they are, an organisation answers for its AI, or an agent acts on someone's behalf, the proof is only worth something if anyone can check it. That takes shared infrastructure, not one more platform.The Verifiable Trust Infrastructure (VTI) is that layer, being specified in the open by Trust over IP and DIF. Our seals, receipts and mandates are designed to plug into it as it matures.

    Read the VTI draft specificationWorking draft 0.2 · not yet ratified
    1. 01Where trust is needed

      People, organisations and their AI

      A person proving who they are, a company answering for its model, an agent acting for someone.

    2. 02What we build

      Trust Seal · Decision Receipt · Agent Mandate

      The proofs we build: a result anyone can verify, a receipt the person keeps, authority an agent can show.

    3. 03Where our founder contributes

      Verifiable Trust Infrastructure

      How people, communities, organisations and agents vouch for each other, delegate and revoke, with no gatekeeper.

    4. 04Open foundations

      W3C DIDs and Verifiable Credentials

      The open identifiers and credential formats everything above is built on.

    01

    Checkable by anyone

    A proof is only useful if a stranger can verify it. Shared infrastructure means no single company decides who is trusted, us included.

    02

    People stay in control

    A person holds their own credentials, vouches for others, and can hand authority to an agent and take it back, without a platform in between.

    03

    Built in the open

    VTI is specified in the joint Trust over IP and DIF Decentralized Trust Graph Working Group, where our founder is a contributor.

    The Validant platform

    One platform, three bodies, every claim auditable.

    Fairness, digital trust and identity run as modules on one platform. Pick what you need, share the evidence, and keep every claim open to inspection.

    What we will not negotiate

    Five commitments, built into the runtime.

    1. Standards before features

      W3C Verifiable Credentials, DIDs and the open work at Trust over IP and DIF come first, with MLflow and pytest hooks for the tools you already run. We extend ecosystems, we do not lock them in.

    2. Rigour a peer reviewer accepts

      Bootstrap and Bayesian intervals, permutation testing, effect sizes, multiple-testing correction. Every claim survives a peer reviewer reading the table.

    3. Compliance is the default

      EU AI Act risk classification, GDPR-aligned privacy tiers and 43 historical discrimination patterns across US, EU and Swiss jurisdictions, wired into every assessment.

    4. Auditable in production

      Real-time drift detection, a fairness gate before release and four-eyes validation. What an agent does on your behalf is logged, signed and revocable.

    5. Privacy and sovereignty by default

      Three privacy tiers per dataset: k-anonymity, differential privacy or exact values. Credentials and mandates are held by the person, never by us. No silent telemetry, no shadow profile.

    Open the platform
    vfairness · open libraryBeta

    The fairness library that powers the platform.

    vfairness is the engineering core inside validant.ai: a Python library for measuring, mitigating, explaining and monitoring fairness across the whole pipeline, LLMs included. Tap any number below for a plain explanation.

    Read the library reference
    News & updates

    Signal.

    What's new at validant.ai — product films, releases and notes from the work. The latest, in order.

    View all
    Featured · Research
    5 October 2026Daniel Glinz15 min read

    Controlling Agents Triples Their Value. Proving It Is What Makes It Count.

    BCG surveyed 1,330 executives and found that companies which control their AI agents properly get about three times as much value from them. Only 5% have those controls in place, and every one of the six is checked by the company itself. A reading of the Applied AI Index 2026, the gap it leaves open, and why proof is what lets the tripled value count.

    Read the report
    ResearchOpen to read
    28 September 2026

    The Preemptive Assurance Paradox: Who Guards the AI Guardrails?

    In one September week a lab asked to slow down, a president declared himself the only guardrail AI needs, and The Economist asked whether the arms race can be stopped. Notes from Trust Valley Days 2026 at EPFL on why the answer is not a pause or a sprint, but liability, outcomes and verification.

    Read
    ResearchOpen to read
    10 August 2026

    Precision Is Not Proof: The Trap in Every AI Fairness Verdict

    A model that predicts perfectly stops insuring anyone; a verdict published without its detection limit invites you to assume the limit is zero. Two failures, one omission, and two borrowed words for keeping them apart.

    Read
    ResearchOpen to read
    24 July 2026

    The Trust Problem Nobody Wants to Name

    Everybody is spending on AI; almost nobody can say what they got back. A new paper, awarded the Best Paper Award at the IEEE Swiss Conference on Data Science and AI, names that gap and shows it is a trust problem wearing a technical costume: seven mechanisms, a four-layer architecture, an iceberg beneath it, and five design principles for closing it.

    Read
    ResearchOpen to read
    14 June 2026

    When a Model Becomes a Munition

    A frontier AI model was switched off worldwide by a single government letter. Reading the June 2026 shutdown through the three-body picture of digital trust: how the kill switch stopped being a metaphor, why the most governable lab was governed least carefully of all, and what single-vendor, single-jurisdiction dependence now costs a board.

    Read
    ResearchOpen to read
    11 June 2026

    Agents Will Act for Us. Who Vouches for Them?

    When agents start acting for us, trust moves from outputs to actors. Reading Michael Casey’s case for proof of control alongside our three-body picture of digital trust: where convenience quietly pulls authority toward a few global platforms, how validant.ai answers it, and a closed beta for scientific validation opening in Q3 2026.

    Read
    ResearchOpen to read
    5 June 2026

    Digital Trust Is an Orbit, Not a Pillar

    Trust is not one more pillar to stack. It is the orbit three bodies trace together: the model, the person and the organisation. Why the three-body problem is the honest metaphor for trustworthy AI, and how to tell where you are in the orbit.

    Read
    EventsOpen to read
    30 May 2026

    Two Views of One Decision: Trustworthy and Explainable AI in Practice at HSLU

    Notes from a Lucerne specialists course on Trustworthy and Explainable AI, and what it confirms about the Validant.ai approach.

    Read
    StudiesOpen to read
    27 May 2026

    The Wrong Question, Asked at Scale

    A landmark study of 4 million job applications shows how AI hiring tools hide their bias, why one rejection can become rejection everywhere, and why independent, position-level assessment is no longer optional.

    Read
    EventsOpen to read
    26 May 2026

    No System Has Ever Been Fair

    What four breakout sessions, one fairness tool demo, and 50+ years of collective experience taught us about fairness in AI, at the Trustworthy AI Circle.

    Read
    UpdatesWatch preview
    22 May 2026

    Two new demo videos: see Pulse run an audit, and the Modules behind every verdict

    We recorded the platform in motion — a full fairness audit start to finish, and the depth behind every result.

    Watch & read
    ResearchOpen to read
    21 May 2026

    Bias is the Foundation

    Why every fairness claim begins with a bias diagnosis, and why skipping it breaks everything downstream.

    Read
    ProductOpen to read
    2 May 2026

    Inside Trust Signal: the weekly newsletter our AI team writes itself

    Every Tuesday, an eight-agent AI team scouts, scores and drafts the week in AI trust and fairness. Here's how it works — and how to get it.

    Read
    Trust Signal · Weekly newsletter

    Get the week in AI trust and fairness.

    An eight-agent AI team scouts 50+ sources, scores every story, and drafts a sharp weekly digest on digital trust, algorithmic fairness and the regulation closing in around them. A human signs off before it ever reaches your inbox.

    • Every Tuesday
    • Free
    • Unsubscribe anytime
    Subscribe to Trust SignalBrowse the archive
    See how it's made
    FAQ

    Frequently asked questions

    What is validant.ai?
    validant.ai is an independent assurance platform for trustworthy AI, organised around three bodies: the model, the organisation and the person. Today it runs AI fairness and explainability audits, and we support digital-trust audits on the iceberg.digital framework as a service while the platform modules are in progress; self-sovereign identity and signed agent mandates are the next body we are bringing into the platform.
    What does validant.ai do for AI fairness?
    It runs end-to-end AI fairness and explainability audits: detecting bias and proxy variables, mitigating disparities across data, models, LLMs and agents, explaining decisions with SHAP, LIME and counterfactuals, and monitoring fairness in production with auditor-grade, regulator-ready reports.
    Who is validant.ai for?
    Anyone who wants AI to stay fair, understandable and under human control. That includes risk, compliance, legal and ML teams proving an AI system is fair, explainable, trustworthy and accountable under the EU AI Act, NIST AI RMF and ISO/IEC 42001, and just as much the everyday person who wants to safeguard their own data and keep control of their agency. We deliberately keep it plain enough for everyone to use, not only specialists.
    What is Verifiable Trust Infrastructure (VTI), and how does validant.ai relate to it?
    VTI is an open specification, currently a working draft, developed in the Decentralized Trust Graph Working Group run jointly by Trust over IP and the Decentralized Identity Foundation. It defines how people, communities, organisations and the AI agents acting for them vouch for each other, delegate authority and revoke it without a central gatekeeper. validant.ai's founder is a contributor at both foundations and active in that working group, and validant.ai's seals, decision receipts and agent mandates are designed to plug into VTI as it matures.