01
Liability
AI underwriting is coming. So is the liability. Banks hire outside auditors for a reason. An independent party stands behind the check.
Open benchmark
A neutral test
for AI underwriting
You are invited to join as a founding member or a founding researcher.
Why a neutral test
02
No neutral test
Every vendor grades its own homework. Today, there is no neutral test.
03
The check
When one goes wrong, you must show it was tested independently.
How the test runs
01 Held out
Same files
Test cases are kept in a separate private store. They are never put in the public set.
02 Same rules
Every tool
Same cases. Same rules. One open contract. Any tool can plug in.
03 Scored
Named result
Each score names the model, the setup, and the lane. Scores are benchmark artifacts, not credit opinions.
To Q2 2027
-
Now
Public alpha. No results published yet.
-
Planned
210 held-out cases, planned.
-
Q2 2027
First public results by Q2 2027 at the latest.
-
After
After that, the rules are set.
Founding seats
Still open
3 to 5
- 3 to 5 founding seats. Each one comes with a board seat.
- Founding members are named in the research, as an independence disclosure.
- Founding members may see their own scores privately before publication. Those scores are locked and can't be altered.
Join
Tell us if you want to join as a founding member, a contributor, or a lender participant.
Or book a call. The booking page is on a SecureLend calendar.
About
SecureLend is one participant and a founding contributor, and it has a conflict.
UWBench was started by SecureLend, which builds AI underwriting tools and takes part as one participant. SecureLend holds one board seat, no veto, and recuses itself on its own scores. Answer keys are held by a neutral custodian.
SecureLend is the first reference participant, not the owner of this repository.