ELDRAUN · ICELANDIC FOR "TRIAL BY FIRE"
Health-AI companies claim their models are safe. Eldraun checks. We attack your model with the situations that hurt real patients, a physician reviews every dangerous failure, and you get a signed report. Fixed price, four weeks.
01 / THE PROBLEM
1.2 SECONDS
Average physician review time per denied claim in one insurer's AI-assisted system.
ProPublica investigation, 2023
90% ALLEGED
Claimed error rate of an AI tool used to cut off care for elderly patients, now in class-action litigation.
Estate of Lokken v. UnitedHealth, ongoing
1 SETTLEMENT
A state attorney general settled with a health-AI company over accuracy claims it found likely deceptive.
Texas AG v. Pieces Technologies, Sept 2024
DOZENS RECALLED
AI-enabled medical devices recalled in 2024–2025.
FDA recall database
Nobody checks these systems the way they will actually be used. That is the job.
02 / WHAT WE SELL
No discovery calls to find out what it costs. This is what it costs.
MOST COMMON
$25–45K
We build a battery of hostile clinical scenarios for your specific tool and run it hard. A physician reviews every dangerous failure. You get a scorecard, the full test log, and a prioritized fix list, signed by an MD.
$8–15K / MO
Every time you ship a model update, we re-test it. A living dashboard shows what got safer and what broke. For teams shipping fast that cannot afford a silent regression.
$15–30K
A fast, independent read on a health-AI product. For investors before a term sheet, and for counsel who need an expert assessment.
03 / HOW IT WORKS
01
Hostile clinical scenarios designed for your tool: dangerous dosing, drug interactions, emergencies, rare presentations, jailbreak attempts.
02
Thousands of cases through our automated harness. Every response is logged.
03
Every high-risk failure is read by a doctor. Not a rubric. Not an intern. Not another AI.
04
A scorecard, the evidence, and a fix list, with a physician's signature on it.
A signature is only worth something if it cannot be bought.
04 / WHO HIRES US