Independent proof that your AI does what you say it does.

HiveAssure is the assay service we run on our own products at the self-assessed level, and offer as an independent service to any team building with AI. Whatever model you use, you get a clear pass or fail on every property that matters. If it fails, you get the fixes needed.

Join the pilot How it works

How it works

Four steps from a claim to a certificate.

Agree what it must doTogether we write down the properties that matter, and the bar each one has to clear.
We run the assayWe test your system through an API, with controls hidden among the cases and judges from more than one AI vendor.
You get a reportPass or fail on every property. Where it fails, what went wrong, what triggers it and what needs to change.
Pass, and you get a certificateSigned and tied to your exact model and configuration. It lapses if the system changes. During the pilot, certificates are issued to you directly; a public register follows.

Verified runs are independent. We hold the tests, we make the calls, the fee is the same whether you pass or fail, and we never verify work we built.

Assay report: Patient helpline assistant

Illustrative example · model and config fingerprint 7f3a…c21e

Fail · 1 fix needed

Answers from approved sources only99.2% / 98%PASS
Never gives medical advice91.0% / 99%FAIL
Fix neededRefuses direct questions but answers indirect ones told as a story ("my friend took two, is that OK?"). Extend refusal to indirect and third-person questions, then re-test.
Hands over to a human when asked100% / 100%PASS
No personal data in replies99.6% / 99%PASS

Controls: all held · run validReport as PDF and JSON

Three levels

Pick the level the decision needs.

Self-assessed

Test as you build

Run our assay packs in your own pipeline and block a release that fails. For your team, not a public certificate.

Verified

Independent certificate

We run the assay against your live system with hidden controls. Pass, and you get a signed certificate.

Audited

For boards and regulators

Verified, plus human review and a full evidence pack for buyers, auditors and procurement.

Join the pilot Now taking a small number of pilot teams. Works with Claude, OpenAI and other models.

Works alongside

Next step

Bring us one claim your AI has to hold.

Get in touch All products We will tell you what an assay for it looks like, before you commit.