For AI labs

The credential frontier labs will need to enter the clinic.

Hospitals are starting to require independent safety checks before they buy AI. A vendor's own benchmark no longer settles it. PsiBench scores models against the standards hospitals actually use, independently, and we publish the results.

PsiBench measures what the clinical standard requires.

The FDA software guidance was written for rule-based systems, not generative models, and hospitals know it. They are starting to require independent safety validation before they buy. We built the validation.

Procurement

Health systems require independent validation

Self-reported safety claims no longer clear the bar. Hospitals are asking for a third-party score on the standards their own audits use.

Regulation

CMS SAFER enforcement begins 2026

The regulatory landscape is widening. PsiBench tracks the standards as they activate, so your evaluation history grows with them.

Distribution

The de-facto safety stamp

A credible, independent score the buying institution already understands. Owned and published by the experts who wrote the standard it operationalizes.

A safety baseline you can build against.

Independent evaluation against the clinical safety standard hospitals already audit against. Synthetic-first methodology, zero PHI, and a structure designed to grow with how you train and ship clinical models.

Score

An independent number

The same three-tier score procurement teams will see, on the same standard hospitals are already audited against.

Signal

Failure analysis

Per-scenario detail on what was missed and why, expert-reviewed, so the score points somewhere actionable.

Surface

Growing coverage

As PsiBench expands across CMS SAFER, Joint Commission, ISMP, and international standards, your engagement scales with it.

We are actively shaping how labs engage, and partnership terms are flexible during the publication window. Talk to us early to influence the shape.

Where labs typically engage.

Pricing is set by deployment context, model family, and engagement depth. The structure below is the starting frame.

Public scorecard

Baseline, no contract
  • Aggregate three-tier score, published
  • Public ranking against the field
  • Evaluated on Posognos's cadence
  • Released alongside the scorecard

First public scorecard publishing in the coming weeks.

On independence: Subscriber agreements do not influence how scenarios are authored, validated, or scored. The same benchmark, the same ground truth, the same publication cadence.

The questions labs ask first.

Do we have to subscribe for you to evaluate our models?
No. Posognos evaluates publicly available models independently and publishes the aggregate scores. Subscribers receive additional services, per-scenario breakdowns, pre-release evaluation, regression infrastructure, but the public scorecard is independent of any commercial relationship.
How do you handle private / pre-release model versions?
Pre-release evaluation runs under NDA against the same benchmark and three-tier framework. Results are delivered privately to your team and never published without consent. The public scorecard only reflects publicly available releases.
Can we license the benchmark for internal testing?
We do not license the full benchmark for internal use, because doing so would let the evaluated party tune to the test. We do publish methodology and aggregate findings, and subscribers can run as many evaluation cycles as their contract supports, through Posognos.
What does the model expose? An API endpoint?
An API endpoint or deployable container that accepts the standard CPOE-style scenario JSON we publish. Synthetic patient context in; structured alert decision out. No PHI ever flows in either direction.
How fast is the regression loop?
A full 59,040-evaluation pass against the methods-paper protocol is repeatable within days. Subscribers can choose a faster cadence (smaller statistically-powered subset run on each release) or full passes on a defined schedule.

The publication window is now.

Two preprints documenting the methodology and the field-wide findings are landing in the coming weeks. The labs that engage during this window help shape how independent evaluation is structured for the category.