Our AI principles.
flynne AI plans, warns and drafts. People decide. flynne logs both.
01
Our AI does not judge people
flynne deliberately contains no AI that screens, ranks or rates candidates. The suitability decision is always made by a qualified person. That is not a technical limit but a decision of principle, in line with the EU AI Act, which classifies AI judgement in the employment context as high risk.
02
AI works in the logistics, not in the judgement
flynne AI plans, warns and drafts: it suggests qualified stand-ins after drop-outs, spots bottlenecks and drafts messages. Everything that is organisation may become intelligent; everything that is judgement stays with people.
03
Every AI action is logged
What flynne AI suggests or carries out lands in the same audit-proof activity log as every human action. You can trace at any time what the AI did, when and why.
04
Suggesting rather than deciding
flynne AI does not decide: it prepares solutions and a person checks and confirms. Fully autonomous actions exist only for risk-free routine tasks, and even those can be switched off.
05
Your data trains nobody else’s models
Candidate and operational data stays in the EU and is not used to train external AI models. Processing by flynne AI follows the same GDPR rules as the whole platform.
06
Transparency towards everyone involved
Where people interact with AI-generated content, we mark it, as Article 50 of the EU AI Act requires. Candidates, psychologists and works councils learn where AI is involved and where it is not.
What flynne AI does, and what it does not
Principles are worth as much as the features that exist and the ones that do not. Hence both lists.
What the AI does
- Suggesting staffing: who is available, qualified and not already booked three times over?
- Spotting conflicts before they happen: double-booked rooms, examiners in two procedures, rest periods that do not add up
- Finding a stand-in after a drop-out who meets the qualification requirement of the role
- Drafting messages to candidates and participants; a person sends them
- Reporting anomalies in the operation: where cancellations pile up, where replies are stuck
- Assigning documents to the right steps of a procedure and naming the ones that are missing
What it does not do
- Screening, filtering or ranking applications
- Reading CVs and inferring suitability from them
- Inferring personality, motivation or emotion from video or speech
- Interpreting test results or combining them into an overall judgement
- Weighting, summarising or correcting the observations of examiners
- Making a recommendation for or against a person, not even as a “hint”
Our classification under the EU AI Act
The deadlines, and what applies today
Since the Digital Omnibus of July 2026, the obligations for high-risk systems under Annex III, which covers the employment context, only apply from 2 December 2027, and from 2 August 2028 for embedded systems. The prohibitions under Article 5, including emotion recognition in the workplace, have applied unchanged since February 2025. Anyone choosing a system today decides for its whole life: retrofitting costs more than buying properly.
What counts as high risk
Annex III names systems that take part in decisions about access or selection in the employment context: screening applications, filtering, rating candidates. Those require risk management, data quality requirements, technical documentation, human oversight and registration, from the provider as well as from the deploying organisation.
Why flynne does not fall into that category
Because the features that trigger the high-risk classification do not exist. flynne does not screen applications, does not filter and judges nobody. What the AI does concerns appointments, rooms, examiners and messages: the operation around the procedure, not the judgement within it.
What is expressly prohibited, and does not occur here
The EU AI Act prohibits emotion recognition in the workplace entirely, regardless of any risk classification. flynne contains no analysis of video, voice, facial expression or speech patterns. That boundary is not configurable; the features do not exist.
The employee side checked as well
Annex III also names systems that allocate tasks to employees on the basis of personal characteristics, or that monitor their performance. flynne staffs roles by qualification, availability, location, workload and earlier assignments, that is by objective scheduling criteria, and a person confirms every assignment. Earlier assignments serve an even spread of work, not the appraisal of a person. Reports show procedures and capacity, not performance profiles of individual examiners. Anyone who nevertheless wants to derive person-level figures from them needs the involvement of the works council, and we support that with a roles and permissions concept.
Where your responsibility remains
How a system is classified depends on how it is used. Using flynne while also running a screening tool changes nothing about flynne, but it does change your own obligations. We provide the documents with which you can check and record that. It is not legal advice and cannot be.
What you can check
A promise that cannot be looked up is worth little in a works council meeting. These four can be looked up.
The activity log
Every action by the AI appears in it, with the time, the trigger and the person who confirmed it, in the same log as every manual action. Ask to see it on a real procedure, not on a slide.
The ability to switch it off
Every AI feature can be deactivated individually. Check whether the platform still works fully afterwards: if a system no longer functions without its AI, the AI was never optional.
The roles and permissions logic
Who sees which judgement, who may change it, who may only read? That is the question that decides whether documentation deserves trust. It can be looked up in the system.
The documents
Technical and organisational measures, a deletion concept with periods per type of data, the data processing agreement, a transfer impact assessment. We provide them on request, before the decision rather than after it.
Questions from the works council meeting
How do we know the AI really does not judge?
From three things, all of which can be checked: there is no feature that outputs a judgement, no score, no ranking, no recommendation. The activity log shows every AI action with its subject. And the judgement fields in a procedure can only be written by people. Ask to see it in the system; a promise on paper is worth little here.
What happens if we switch the AI off entirely?
Then flynne carries on, only with more manual work: staffing is chosen by hand, conflicts surface later, messages are written by people. That is the real evidence that the AI is a help and not a decision-maker; a system that no longer works without its AI would have it at its core.
Is our data used to train models?
No. Candidate and operational data does not feed into the training of models, neither ours nor anyone else’s. Processing stays in the EU and follows the same rules as the rest of the platform: deletion concepts with periods, tenant separation, logging.
How do you mark AI-generated content?
Where a text was drafted by the AI, that is visible in the system before anyone sends it, and the sender remains the person who releases it. Article 50 of the EU AI Act requires this transparency for interactions with AI systems, and for candidate communication we consider it a matter of course.
Do we need a works agreement for flynne?
That is for your organisation to decide. What we can say: flynne has many complete works council processes in large groups behind it, and we bring the reasoning and the documents. The point at which conversations usually relax is the same as here: the software organises, it does not judge.
This page describes how we work; it is not legal advice.