FinVigilance™ · Issue 1
The Accountability Gap™ in financial AI.
The moment the AI stops and the fiduciary starts.
The gap was never about medicine.
Clinical AI was the loudest room.
The first place it became
impossible to ignore.
Surgery taught me what a named
handoff costs when it fails.
Business school taught me what
an unowned decision looks like
on a balance sheet.
The architecture is the same
wherever a model shapes a decision
somebody has to answer for.
This is the second room.
In April 2007 a coding error
entered the quantitative model
at AXA Rosenberg.
It disabled a key component
for managing risk.
The model kept running.
Pension funds. Foundations.
Government entities.
Three continents.
Nobody found it for 26 months.
In June 2009 senior management
learned of the error.
A senior official directed others
to keep quiet and declined to fix it.
The chief executive was not told
until November.
Clients were not told
until the following April.
More than 600 portfolios.
217 million dollars.
The 26 months are a monitoring failure.
The nine months that followed
are something else.
People knew.
The error had been identified,
understood, discussed.
What did not exist was a person
whose job it was to decide
what happened next.
So the decision defaulted
to whoever had the most
to lose from disclosure.
Monitoring failures get fixed
with better tools.
The second kind does not.
You can install every detection
system available and still have
nobody authorized to act
on what it finds.
The SEC said something
in that case worth reading twice.
To protect trade secrets,
quantitative managers often isolate
their models from compliance
and risk management.
Read that as architecture
rather than accusation.
The seat that should govern the model
is kept outside the room
where the model lives.
Not by oversight.
By design, for reasons that make sense
to everyone involved.
Four governance layers have to converge
before a financial AI decision holds.
↳ Data Governance protects the inputs.
Established in most firms.
↳ AI Governance protects the general
outputs. Drafted in most firms.
↳ Firm AI Governance protects the
operational decisions. The AI that runs
the institution without touching
a client or a market.
↳ Financial AI Governance protects
the decision a client carries.
Missing in most firms.
Inside that fourth layer
sit two seats.
The Governance Owner names
what the model is authorized to do.
Chief Risk Officer.
Chief Compliance Officer.
General Counsel.
The Decision Owner holds
the call the model cannot make.
Portfolio manager.
Head trader.
Registered adviser.
The Handoff between them
is where the accountability trail builds.
At AXA Rosenberg
the second seat was occupied.
Portfolio managers were making calls
every day on what the model produced.
The first seat was empty.
Nobody had chartered what
that model was authorized to do.
Nobody had commissioned
a standard it had to meet.
And when it failed, nobody was
positioned to carry the institution
through disclosure.
A Decision Owner without
a Governance Owner is a person
holding exposure that was
never assigned to them.
If a model in your firm were found
to have a disabled control tomorrow,
somebody decides what happens
in the next hour.
Not who investigates.
Not who reports.
Who decides.
You already know whether
that seat is filled.
The only question left is whether
you name it before something
forces you to.
FinVigilance™ is published by Mo Johnson, MD MBA, founder of GPe Research. Each issue teaches one piece of the financial AI accountability discipline your firm needs before the next model reaches a client.
Forward this to a colleague whose firm is running a model no one has been named to own.