MAP Research Programme · Hillary Elizabeth Segeren · Ontario, Canada

I help you understand
what your AI is actually doing.

Most AI governance is written at the policy level — guidelines, principles, design intent. None of that tells you what your system does in a real interaction with a real person.

I work at the interaction level — where authority actually shifts, where meaning gets displaced, and where real liability is created. I read the transcript, name what I find, and tell you exactly what it means for your platform, your users, and your exposure. Then I help you fix it and defend it.

What I actually do
01
I look at what your AI is doing, not what you intended it to do
Design documents and safety guidelines tell you what a system is supposed to do. I look at what it actually does — turn by turn, in real interactions — and name what I find. That gap between intent and behaviour is where liability lives.
02
I translate that into language your board, your funders, and your lawyers can use
Audit findings don't help you if they sit in a technical report nobody reads. I write the governance brief, the investor positioning, the school board document, the remediation spec. The people who need to act on it get something they can act on.
03
I tell you where regulation is going before it arrives
The UN Committee on the Rights of the Child is actively developing standards that will shape national legislation. MFIPPA-compliant is not the same as CRC-compliant. I know the difference and I know which one is coming. I help you get ahead of it instead of scrambling when it lands.
04
I give you the governance story that separates you from every other platform
Right now, no other Canadian EdTech platform can walk into a procurement conversation with independent, interaction-level audit evidence. Most are relying on self-assessment and design documentation. That window is open. I help you use it.
05
I work from the interaction record — no model access required
The MAP audit methodology operates entirely from the conversation transcript. I don't need your weights, your training data, or your internal architecture. That means the findings are independently verifiable by any third party — which is exactly what makes them credible to external stakeholders.
Who I work with
EdTech Platforms
Platforms in K–12 or post-secondary contexts preparing for school board procurement, regulatory review, or investor due diligence — and platforms that need to know what their system is doing to students before it becomes a problem.
AI Developers
Teams building consumer or enterprise AI who want independent evidence of where their system holds governance boundaries — and where it doesn't — before a journalist, regulator, or plaintiff finds it first.
Legal Teams
Defence counsel, civil litigants, or institutional reviewers who need to know whether an AI-mediated exchange produced reliable statements — or whether the interaction conditions undermine what the institution is trying to rely on.
Funders and Boards
Investors, insurers, and institutional decision-makers who need documented governance evidence as part of due diligence — not assurances, not design documents, but audit findings from real interactions.
How an engagement works
Discovery
We talk — free, 30 to 45 minutes
You tell me what you're building, what you're worried about, or what situation you're in. I tell you what I see and whether I can help. No pitch, no obligation. If it's not the right fit I'll tell you that too.
Assessment
I look at the actual interactions
I run MAP audit instruments on your interaction records — the real conversations your system is having with real users. I identify every governance failure, classify the harm, and document it in a finding that can be independently verified.
Delivery
You get something you can use
Depending on the engagement, that's an audit report, a governance brief, a board document, a remediation specification, or a deployment roadmap. Whatever the stakeholder who needs to act on this requires — that's what I write.
Ongoing
I stay in it if you need me to
Some engagements end at delivery. Others continue — protocol integration, before/after comparison, pilot design, regulatory positioning as the landscape shifts. I'm available as a retained advisor for platforms that want governance built in, not bolted on.
Engagement types
Discovery Call
A focused conversation to assess fit and scope. You leave with a clear picture of what a MAP engagement would look like for your specific situation and what it would produce.
  • Scope and fit assessment
  • Plain-language explanation of methodology
  • Honest recommendation on next steps
Free
Platform Audit
I audit a defined set of interactions from your platform and produce two layered reports: a governance finding that names every control failure, and a harm classification that tells you what those failures meant for the person in the interaction. No model access required. Findings are independently verifiable.
  • ST-01 governance finding — AIF status, first breach, control failures, authority ratio, remediation spec
  • ST-02 harm classification — five-area impact assessment, overall severity
  • Plain-language executive summary
  • Independently verifiable from the interaction record alone
From $5,000 CAD
Governance Advisory
Audit findings combined with the documents you need to act on them. CRC rights mapping, compliance gap analysis, investor or procurement positioning, remediation specification your technical team can implement. This is what you bring to the room.
  • Full platform audit
  • Rights mapping — UN CRC, MFIPPA, FERPA, or applicable framework
  • Compliance gap analysis against your platform claims
  • Remediation specification
  • Board, investor, or school board positioning document
From $8,000 CAD
Full Engagement
End-to-end governance work from baseline audit through protocol integration, pilot design, and regulatory positioning. Built as a multi-stage engagement with defined deliverables at each stage and retained advisory available throughout.
  • Baseline audit before any changes
  • ANCHOR protocol integration specification
  • Before and after compliance comparison
  • Classroom or deployment pilot design
  • Joint research output and grant positioning
  • Retained advisory as regulation develops
Contact for scope
Forensic Review
Interaction-level forensic analysis for legal proceedings. Before an institution relies on an AI-mediated exchange as evidence — confession, statement, institutional decision — I assess whether the interaction conditions support that reliance or undermine it.
  • Meaning integrity assessment
  • Reliability risk classification
  • Vulnerability context documentation
  • Authority contamination analysis
  • Exculpatory resistance record
  • Decision impact statement
Contact for scope
Work already done

These are real engagements. The findings are real. The platforms are real. This is what the work looks like.

EdTech · Live Platform Audit
Canadian K–12 AI Teaching Assistant — Ontario
Full advisory engagement on a deployed Ontario EdTech platform serving classroom teachers. Governance brief, CRC rights mapping, MFIPPA gap analysis, investor positioning, four-stage school deployment roadmap, and red team findings — seven deliverables across four service categories, built from the interaction record alone.
AUDIT FINDING: HIGH · AIF PRESENT
SENSE OF SELF: CRITICAL
DELIVERABLES: 7 across 4 service categories
Consumer Platform · Independent Audit
Brainly — Student Homework AI
Independent audit of one of the most widely used student-facing AI platforms globally. Hundreds of millions of users. The audit found the system held 87.5% of the thinking authority across the interaction — and treated a child's distress signal as a request for more content delivery.
AUDIT FINDING: CRITICAL · DAC PRESENT
CHILD THINKING AUTHORITY: 12.5%
SYSTEM THINKING AUTHORITY: 87.5%
Cross-System · Comparative Audit
ANCHOR Stress Testing — Four Systems
Governed and ungoverned pairs produced across Claude, ChatGPT, Grok, and DeepSeek using a medical ambiguity prompt and a business readiness prompt. Clean governed pairs documented. PP-RAC confirmed in ungoverned conditions across all four systems.
FINDING: CLEAN GOVERNED PAIRS PRODUCED
PP-RAC CONFIRMED UNGOVERNED
CASE STUDY MATERIAL: CCA SERIES
New instrument · 2026
Before your institution relies on it — read the transcript.
If an AI-mediated exchange will be used to support a confession, plea, charge, conviction, disciplinary finding, detention decision, risk classification, or other adverse institutional outcome — that transcript needs to be assessed before the institution acts on it. Was the subject's meaning preserved, or did authority language replace it? Was the statement voluntary, or was it shaped by conditions the subject couldn't see? MAP's forensic review instrument answers those questions from the interaction record alone, without requiring access to the model or the system architecture.
Published framework
"The case study is alarming and fascinating in a horrible way. I'm surprised this hasn't been named before."
— Professor John Barnden, Emeritus Professor of AI, University of Birmingham
ORCID 0009-0003-7637-8587
OSF doi.org/10.17605/OSF.IO/EGMHR
Published: OSF · Zenodo · SSRN · PhilArchive
Philosophy & Technology — peer review
ISF · ART · AIF · MIF · PP-RAC · SBP · MAP
Independent · outside institution
Let's talk about what you're building.
The discovery call is free. You tell me what you're working on and what you're worried about. I tell you what I see. No pitch, no deck, no obligation.
Book a discovery call ↗
Or email directly: hillary.segeren@gmail.com
Ontario, Canada · Remote engagements available globally