The model cannot score.
25.3 million people
have a credit file.
The file exists.
The score does not.
CFPB technical correction.
June 2025.
The old figure said 26 million invisible.
The correction cut it in half.
The unscored population grew instead.
Ten years of lending policy
ran on a number
nobody owned checking.
AI can score them now.
That is the good news
and the entire problem.
A model that scores someone
the system could not score
is not reading a thinner file.
It is deciding
what counts as evidence.
Someone chose which alternative data enters.
Someone decided rent counts.
Someone decided it counts as much
as a closed auto loan.
None of those are credit decisions.
They are authorizations.
And the threshold is the part
nobody can validate.
A cutoff on a scored population
is set against decades
of observed repayment.
This population has none.
The number gets chosen
from a comparable group,
which is another way of saying
it gets borrowed from people
who are not these people.
What the institution sees.
↳ A model that expands access
↳ A population finally reachable
↳ An approval rate that improves
What the file actually holds.
↳ A threshold validated against somebody else
↳ An evidence standard nobody ratified
↳ A denial reason nobody can produce
Expanding who gets evaluated
is a governance decision
wearing a growth number.
The Governance Owner names
what the model may treat as evidence.
The Decision Owner holds
the call on the applicant in front of them.
The Handoff is where
the authorization gets recorded.
The Accountability Gap™ (TAG™)
does not open at the denial.
It opens at the decision to score
someone the system was never built to score.
You already know which model in your stack
reaches a population nobody
formally authorized it to reach.
That is not expanded access.
It is unauthorized underwriting
with a better approval rate.
Who authorized your model
to decide what counts as evidence?
Mo Johnson, MD MBA is a cardiothoracic surgeon and the founder of GPe Research. Field Notes are short dispatches from the financial AI accountability frontier, published alongside the FinVigilance™ newsletter at finvigilance.org
Follow the work on LinkedIn: linkedin.com/in/mo-johnson