Human in the Loop: independent human testing for your AI

Why people, and not the vendor

A vendor cannot judge its own biases. Its governance tool exists to reassure you about its product. A regulator, an insurer or a large client will ask who, outside the vendor, has checked.

The Cockpit's automatic measurements tell you what an AI system costs and who uses it. They do not tell you whether an answer is right, biased or dangerous. For that, you need people who read.

What the offer covers

• Black-box testing campaigns. Bias, hallucinations, data leaks, procedures not followed: scenarios written for your use cases, run against your AI systems as they actually run.
• Two independent reviewers. Each case is judged by two independent people. When they disagree, a third opinion decides.
• Review of real conversations. Samples reviewed by people residing in the European Union, in teams formed for each client to our requirements, within the scope set in the purchase order.
• Test sets for your agents. Building and annotating the scenarios that then serve as your regression tests.

How a finding becomes evidence

Every finding enters the Cockpit as dated evidence, with its author. That is what an inspector accepts, and what no vendor tool can produce about its own AI. The evidence is attached to the systems in the AI register and to the obligations it documents.

Our partner: Isahit

Isahit is a French data-annotation company. The first European ethical-AI company certified B Corp: 1,000+ annotators, 7,500+ projects, references such as Airbus, Orange and L'Oréal.

The offer is activated on engagement: each campaign is prepared for one client, on its use cases, within the scope set in the purchase order.

Building and controlling stay separate

A Human in the Loop campaign never tests a system the Factory built or maintains for the same client. That is what makes the result count as an independent review.

To see where this offer fits, read AI governance.