Responsible AI

#Why principles are not enough

Almost every provider has AI principles. What counts is whether they reach the product: as a setting nobody can forget, as a log anyone can read, as an approval someone is accountable for.

That is why Univerx describes each principle together with the control that implements it. Where a control is configurable, we say so. Where it is a goal, we say that too.

#Five principles, five controls

#How the controls apply

#AI agents

AI agents resolve requests on their own and act in connected business systems while doing so. They identify themselves as AI, hand over to a person with context at any time and log every action.

#Univerx Assist

Univerx Assist suggests, the agent decides. Summaries and reply suggestions rely on the approved knowledge base, not on a model's general knowledge of the world.

Univerx Knowledge

Question

Where is our data processed?

In EU regions of your choice. Hosting in Germany is planned. The sub-processor list and the processing-agreement documents are available on request.

SourceTrust Center · Data protection

#Semantic²

Semantic² reads documents and reports discrepancies as findings in euros, with rule and source. No finding is booked automatically. A person makes every decision, and it is logged.

#Governance day to day

Governance is not a committee that meets once a year. It is a process every change to an AI goes through, from first draft to rollback.

1. Design. A team builds or changes a flow in Agent Studio, without code.
2. Review. The team checks the flow against examples before it reaches customers.
3. Approve. An accountable person approves model and prompt; the version is stored.
4. Observe. Conversation analytics and logs show how the flow performs in operation.
5. Roll back. If needed, an earlier version is restored.

#Read on

How the principles relate to the duties of the AI Act is set out on EU AI Act. The guide Introducing AI agents responsibly turns the points to settle before launch into a checklist.

Why a well-kept knowledge base is the foundation of grounded answers is the subject of Grounded answers. The product page Responsible AI gives an overview of the features.