amilieu

Safeguarding our Future

AI safety is the biggest issue of our time.

Amilieu was created to serve this critical need. We keep watch over AI once it is in use: a platform that monitors how each system behaves, and an oversight team that works with the people who use, supply and answer for it.

Health · Education · Defence

How it works

How it works

Software that watches.
People who act.

01 · The platform

Every AI system, monitored continuously.

Register each system, keep its evidence in one place and track how it behaves in use: accuracy, hallucination, drift and incidents.

See the platform

02 · The oversight service

People who understand the setting.

We work out who each system affects, agree with them what to watch, and follow every finding through to a decision.

See how oversight works

The platform tells the team where to look. The team decides what the platform should watch.

The gap

Most AI safety work
stops at launch.

AI systems are tested before they are bought and approved. Then they change. Vendors ship updates, new users and data arrive, and staff find new ways to use them. Very few organisations have anyone watching for it.

One AI system, from purchase to years in use

Without ongoing oversight, each change after launch moves the AI system further from what was approved, and nobody notices. With Amilieu, checks continue after launch, each change is caught and reviewed, and the system stays close to what was approved.

Go-live

What you approved

Safety checksNo checks after launch

The gap. Nobody is watching it widen.

  1. 1

    Vendor update

    The model you approved is not the model you run.

  2. 2

    New users

    It meets people and cases it was never tested on.

  3. 3

    Inputs drift

    The data it sees slowly changes.

  4. 4

    New uses

    Staff find workarounds and new tasks for it.

Schematic illustration, not measured data.

Part one · The platform

Every AI system,
in one place.

A shared record of each AI system you run: who owns it, what evidence supports it, and how it is behaving right now. Built for the people who answer for AI, not only the people who build it.

amilieu.io / governance

Governance Workflows

Manage compliance, approvals, change control

live

Checklists

10

Approvals

10

Change Req.

4

Eval plans

10

APP1110Approvals10Eval10+ Request

Evaluation plans

10 total · 2 running · 1 queued

Echo-Cardiac v3.2

Plan 6/6 · clinical envelope

Passed

Triage-ED-2

Plan 2/6 · cohort match

Running

Ambient Scribe v2.1

Plan 0/4 · awaiting sign-off

Queued

PathAI-Histology

Plan 12/12 · equity audit

Passed

Sepsis Early Warning v3.1

Clinical review · production

Approved

Chest X-ray Triage AI

Clinical review · production

Approved

Head CT Haemorrhage Triage

Clinical review · production

Approved

Stroke LVO Detection AI

Clinical review · shadow

Approved

Discharge Readiness Score

Pending lifecycle review

Pending
Avg review: 2.3 days · Median approval: 1.7 daysUpdated 12s ago

01 · Govern · Health

Know what you run and who owns it

A register of every AI system: its purpose, risk level, accountable owner, review dates and audit trail.

amilieu.io / registry / lesson-planner

Lesson Planner

School system · lesson planning

HealthyProductionMedium riskTeacher review

Accuracy

98%

Accuracy

98%

vs baseline

Open alerts

1

0 critical

Open tasks

0

Nothing pending

Ingestion

12h

0.0% fail

Quality1ActivityCompliance & Config

Recent activity

live · 24h

12:34Drift snapshot captured
12:18Sign-off received · teaching lead
11:52Review set added · 240 lessons
11:30Curriculum review completed

Hallucination rate

live · 30d

02 · Assess · Education

Test it against your setting

Evaluate each system against its intended use and your own data before it goes live, and again when it changes.

amilieu.io / defence / monitoring

Defence monitoring

AI Safety & Oversight Platform

all audited

Active

10

Incidents

4

Reviews

8

System health

Healthymedium

Misinformation flags

Analysis team A

98%

Healthyhigh

Source verification

Open-source review

93%

Healthymedium

Narrative tracking

Analysis team B

87%

Healthyhigh

Claim matching

Source comparison

92%

Monitoring alerts

ops.ingest_failure

high

False-positive cluster · source set B

slice.performance

high

False positives · source set B

drift.psi

high

Source distribution changed

drift.psi

warn

Unsupported claim rate

calibration

warn

Confidence drift · claim matching

2.3k events/min · 99.99% uptime · 1 in 7dSynced 12s ago

03 · Monitor · Defence

Catch drift the day it starts

Continuous monitoring for accuracy, hallucination and drift, with thresholds agreed with the people who use the system.

Tell us which AI systems you run and we’ll walk you through the platform set up around them.

Illustrative views with example data.

Part two · The oversight service

AI-assisted monitoring.
Human-led oversight.

AI helps us monitor at scale and surface potential issues. People with knowledge of the setting define what needs review, interpret findings and decide what action to take. Automation supports that work; responsibility stays with people.

Choose a sector above, then hover over or tap a group below to see its connection and question highlighted together.

CliniciansExecutivesAI vendorEHR vendorRegulatorsPatients

  1. 01

    Map

    Identify everyone the system affects, inside and outside your organisation.

  2. 02

    Agree

    Settle with each group what to watch, what counts as a problem, and who responds.

  3. 03

    Follow through

    Route findings to the people who can act, and keep a record of every decision.

Ways to work with us

Start where
you are.

Buying an AI system, selling one, or running many: each calls for a different kind of oversight.

01 · For organisations buying AI

AI system check

An independent check before you commit.

We test a system against how you intend to use it, examine the vendor’s evidence, and give you a clear finding to take into the purchasing decision. Comparing products? We check them side by side.

  • A written assessment in plain language
  • Evidence gaps and questions to put to the vendor
  • Conditions for safe use, if you proceed

One-off · fixed scope

CANDIDATESAMILIEU CHECKFINDINGSSystem AProceedSystem BWith conditionsSystem CNot yetTESTED AGAINSTIntended useVendor evidenceYour settingBEFORE YOU SIGN

02 · For AI companies

Vendor partnership

Independent assurance that travels with your product.

Hospitals, schools and government buyers want someone other than the vendor checking the product. We monitor it independently at each customer site, share findings with both sides, and help you meet each new customer’s requirements.

  • Independent monitoring at every customer site
  • Evidence your customers’ governance teams can use
  • Support entering new organisations and sectors

Ongoing · grows with your customer base

More on partnerships →
VENDORYour AI productDEPLOYED TOCustomer 1Customer 2Next customerFINDINGSASSURANCEAmilieuINDEPENDENT

03 · For government and enterprise

Enterprise & government oversight

Start with one system. Extend to all of them.

We begin with a pilot: one AI system in one setting, with measures agreed with your operational, privacy and compliance teams. Once it has shown its value, oversight extends across your AI portfolio as an ongoing service.

  • A bounded pilot with agreed success measures
  • Continuous monitoring with human review
  • One view of every AI system you run

Pilot, then ongoing subscription

PILOTREVIEWONGOING OVERSIGHT1 system · 1 settingEvery AI system you run

Standards & evidence

Governance,
with evidence.

Know who is responsible, what has been checked and what happens next. We help your team keep that evidence together, with the scope agreed for each engagement.

  1. 01

    Define the use and owner

    Record what the AI is for, who is accountable, and the data, systems and actions it is authorised to access.

  2. 02

    Agree the checks

    Scope the relevant policies and controls with your team. Identify evidence gaps, evaluation criteria and the findings that need human review.

  3. 03

    Monitor and review changes

    Assess performance against agreed criteria and review material changes. Monitoring depth depends on the data and access available for each system.

  4. 04

    Keep the decision trail

    Connect findings to the people responsible, the action taken and the next review. Your organisation retains authority over approval and risk acceptance.

Frameworks relevant to your setting

References such as ISO/IEC 42001 and the NIST AI Risk Management Framework help frame the governance conversation. The requirements and evidence to assess are agreed for your use case; this list is not a completed compliance mapping.

Across sectors

ISO/IEC 42001:2023 · AI management systems · ISO

AI Risk Management Framework 1.0 · NIST

Guidance for AI Adoption · National AI Centre

EU AI Act · Regulation (EU) 2024/1689 · European Union

Government

Policy for the responsible use of AI in government, v2.0 · Digital Transformation Agency

Health

Regulation of software-based medical devices, including AI · TGA

AI Clinical Use Guide · Australian Commission on Safety and Quality in Health Care

Education

Australian Framework for Generative AI in Schools · Department of Education

Defence

A Method for Ethical AI in Defence · Defence Science and Technology Group

Amilieu provides evidence for oversight decisions. An assessment does not certify safety, establish regulatory approval or replace your organisation’s obligations.

Hosting, data & assurance
Hosting options
Deployment options include a dedicated Microsoft Azure instance in Australia East (Sydney), or an organisation’s approved environment. The hosting boundary, services and data location are agreed for the engagement before production use.
Cloud-provider assurance
Microsoft publishes Azure SOC 2 Type 2 audit information. Its coverage applies to the services and controls in the report’s scope; it is not an Amilieu SOC 2 report or certification.
Controls agreed before deployment
Scope who can access data, where evaluation runs, what is retained, what is logged and what security evidence your team needs. Required integrations and assurance checks form part of the deployment acceptance criteria.
Website privacy
Our privacy policy explains personal information collected through this website. Platform data-handling arrangements are defined separately for each engagement.

Our founders

The minds behind
Amilieu.

Clinicians, researchers and technologists building independent oversight for AI.

Daksh Tyagi

Daksh Tyagi

Co-founder & Chief Safety Officer

Read bio
Gihwan Song

Gihwan Song

Co-founder & Chief Technical Officer

Read bio
Joshua Kovoor

Joshua Kovoor

Co-founder & Chief Executive Officer

Read bio
Aashray Gupta

Aashray Gupta

Co-founder & Chief Operating Officer

Read bio

Frequently asked

Answers to the
hard questions.

About the platform, the oversight service, and how we handle your data.

Amilieu is an independent AI safety company. We monitor AI systems once they are in use, and work with the people around each system to interpret what we find and act on it.

No. Our founders began in clinical AI, and the same approach now covers education, defence and other regulated sectors.

Many vendors monitor their own products well. An independent check gives you, your board and your regulators evidence that does not come from the party selling the system. We also involve the people a vendor’s dashboard leaves out: frontline staff, governance teams and those affected by the system’s outputs.

Yes. Most organisations start with one system in one setting, either as an AI system check before purchase or as a pilot of ongoing oversight.

No. It is an independent assessment to inform your decision. Amilieu does not currently issue certifications.

Deployment options include a dedicated Microsoft Azure instance in Australia East (Sydney), or your organisation’s approved environment. Hosting, access, evaluation and retention arrangements are agreed for each engagement before production use. Azure’s provider-level audit coverage does not certify Amilieu or an individual deployment.

No. We help organisations meet what their regulators and their own governance require. We complement regulation; we don’t replace it.

Book a demo

See the platform.
Meet the people behind it.

Tell us a little about your AI and we’ll show you how Amilieu would oversee it. We reply within two business days.

EMAILhelp@amilieu.io
LINKEDINlinkedin.com/company/amilieu
ADDRESS207 / 102 King Street Newcastle NSW 2300

By sending you agree to our Terms and Privacy Policy.

Book a demo

See the platform. Meet the team.

We’ll show you the platform and talk through what oversight could look like for your AI. We reply within two business days.

Prefer email? help@amilieu.io. By sending you agree to our Terms and Privacy Policy.