Responsible AI
The principles we follow when we design and introduce AI.
AI makes mistakes, and introducing it changes how people work. So on every project we settle in advance who is accountable for the result, what the system may do on its own, and how an error is spotted and put right. These principles apply to the solutions we design for clients and to the tools we use while doing it.
On this page
Principles
Seven principles
- People decide
- We distinguish between reading information, preparing a recommendation, carrying out an approved action and a step a person must decide. Where money, external commitments or people are concerned, the final word rests with someone who has the authority to change the decision.
- Controls match the risk
- One rule does not fit everything. Before go-live we assess the impact of an error, the uncertainty of the result and the options for putting it right. Permissions, approvals and monitoring follow from that.
- Data only within the agreed scope
- A system gets access only to what the task requires. For each solution we describe hosting, storage, model services, processors and any transfers outside the EU. We do not use client data to train models unless expressly agreed otherwise.
- Open about what is AI
- People should know when they are working with AI and when they are reading its output. Estimates and recommendations show their inputs and limits. We do not present an illustrative example as a result or a design as a live operation.
- We test before go-live and after
- We test normal cases and exceptions: a missing document, a duplicate, a changed input, an unavailable system. After go-live we track errors and deviations and adjust the rules accordingly. It is clear in advance who can stop a solution and how to return to a fallback procedure.
- Change with people, not around them
- The people affected help shape the change and get support during real work. Solutions that touch on assessing people, such as recruitment or appraisal, get a separate review with the client’s leadership and responsible staff.
- Every solution has an owner and a record
- The client sets the business rules and approves selected steps. ENTER is accountable for the contracted delivery and agreed operation. We agree what is recorded, who may review records and how incidents are handled.
Limits
What we don’t do
We don’t take part in these solutions, even where they would be technically possible.
- Practices prohibited by Article 5 of the AI Act, such as manipulative techniques, social scoring or emotion recognition in the workplace.
- Covert monitoring of employees.
- Automated final decisions about a person, such as hiring or dismissal, without the possibility of human review.
- Passing AI off as a person where the other party has a right to know.
- Claims about results that we can’t evidence.
- Working with data the client has no right to use.
In practice
How the principles shape a project
Audit
We list the systems in use, their purpose, data and responsible people. We identify where the risk is higher and who decides on the client’s side.
Adoption
People learn to check outputs, work within company rules and recognise what doesn’t belong in AI. This also supports AI literacy.
Integration
We set up permissions, approvals, records, monitoring and a way to stop the solution. Only then does it go live.
Regulation
The AI Act and other rules
Obligations depend on what a system does and the role your organisation plays. We help with the technical and organisational assessment. Legal classification needs to be confirmed for each specific use. The Security page covers the timeline and how data is handled.
Depending on the task and contractual requirements, we can work with models from Microsoft, Anthropic and OpenAI. For questions about these principles or to raise a concern, write to milo@enterai.cz.
Version 1.0, effective 30 October 2026. We review these principles at least once a year. Specific commitments for an engagement are set out in the contract.