Conditions for the artificial intelligence features
What the AI features do, what we guarantee, and what you commit to as the organisation using them.
Version 1.0. The Romanian text is the binding one; the English translation is provided for information.
Version 1.0 · in force from the date of publication
This document forms part of the Partnerd Hire Terms of Use. It describes what the artificial intelligence features in the platform do, what we guarantee and what your obligations are as an organisation that uses them in hiring decisions, under Regulation (EU) 2024/1689 on artificial intelligence ("the AI Regulation").
1. The parties' roles
1.1 Partnerd is the provider. Partnerd develops and makes available the artificial intelligence features in the platform and is responsible for their design, documentation and monitoring.
1.2 The Client is the deployer. The Client uses these features in its own recruitment processes, under its own authority, and is responsible for how it integrates them into its decisions.
1.3 High-risk system. Because some features support the evaluation of candidates, Partnerd Hire is designed and documented as a high-risk artificial intelligence system within the meaning of Annex III to the AI Regulation. The obligations the Regulation sets for such systems become applicable from 2 December 2027; Partnerd is implementing them progressively from now.
2. What the artificial intelligence features do
| Feature | What it does | What it does not do |
|---|---|---|
| CV reading | Turns a CV into a structured record: experience, education, skills, languages, certifications | Does not evaluate the candidate |
| Match score | Calculates an indicative score (0–100) between an application and the job requirements, with a confidence level, arguments for and against, and the evidence from the CV it relies on | Does not reject, advance or rank applications automatically; does not receive age, graduation year, address, native language or salary |
| The Nerdy assistant | Answers questions about your candidates and jobs, searches the candidate base, builds shortlists on your request, drafts texts | Does not move candidates between stages, does not reject and does not hire; you confirm any proposed action |
| Text drafting | Proposes job postings, messages to candidates, feedback | The text is edited and sent by you |
| Candidate base search | Turns a natural-language request into filters you can see and change; semantic search orders by closeness to the description | Does not make decisions |
| Candidate coach | At application time, asks the candidate questions to clarify their experience against the job; the answers go into their record | Does not reject the candidate |
| Video interview (Introvid) | Transcribes the recorded answers and evaluates them strictly on content: relevance, skills mentioned, clarity | Does not analyse face, voice, emotions, accent or any other biometric characteristic |
The list of features, the versions of the models used and the detailed instructions for use are documented by Partnerd and made available to the Client on request.
3. What Partnerd guarantees
3.1 A human decides. No artificial intelligence feature in the platform makes a decision with effect on a candidate. Rejection, advancement and hiring are actions available only to human users with the appropriate role.
3.2 Transparency. Every score is shown together with the explanation and evidence it relies on, its confidence level and a warning when the model found insufficient evidence. Every result is visibly marked as generated by artificial intelligence.
3.3 Log. Every use of an artificial intelligence feature is recorded automatically: what was asked, what was answered, which model and instruction version was used, when, and what the subsequent human decision was. The log is available to the Client for its own applications and underpins the explanations given to candidates.
3.4 Reducing the risk of discrimination. The scoring model does not receive fields such as age, graduation year, home address, native language or salary, is instructed to ignore protected characteristics and not to use the writing quality of the CV as a signal. Sensitive data detected in CVs is masked and excluded from evaluation. Partnerd periodically tests the evaluation features for accuracy, stability and disparities between groups and keeps the results.
3.5 Control. The Client can disable the artificial intelligence features for the whole workspace at any time (master switch), can disable automatic scoring and can limit which actions the assistant is allowed to propose.
3.6 No training. The Client's and the candidates' data is not used to train models, neither by Partnerd nor by the model provider.
3.7 Monitoring and incidents. Partnerd monitors the system's operation after it is placed on the market and reports serious incidents to the authorities under the AI Regulation. The Client is informed of any incident affecting its candidates.
4. The Client's obligations
By using the artificial intelligence features, the Client undertakes:
4.1 To use them as intended. The features are support tools for recruiters. The Client does not use them to make automated decisions about candidates, does not build processes in which the score alone determines rejection or advancement, and does not use them for purposes other than recruitment.
4.2 To ensure human oversight. The Client designates persons with the necessary competence and authority, who understand how these features work and what their limits are, who read the evidence before relying on a score, who can ignore or contradict the results and who actually make the decisions. The Client ensures that its users go through the instructions for use provided by Partnerd.
4.3 To inform candidates. The Client informs candidates, before evaluation, that artificial intelligence tools are used in the recruitment process, what they do and that the decision is made by a person. The platform provides standard texts, which the Client may adapt; the responsibility for informing lies with the Client.
4.4 To answer candidates. The Client answers candidates' requests for an explanation of the evaluation assisted by artificial intelligence and for its review by a person, using the tools in the platform.
4.5 Not to enter irrelevant or prohibited data. The Client does not add discriminatory criteria or special-category data to the job requirements or the application questions and does not attempt to circumvent the system's safeguards.
4.6 To report problems. The Client informs Partnerd without delay if it observes systematically wrong or biased results or unexpected behaviour of the features, at hello@partnerd.ro.
4.7 To keep the logs. The logs generated by the platform remain available to the Client for the term of the contract; the Client exports them if it needs them after the contract ends.
4.8 Employers in other countries. If the Client uses the platform for recruitment outside Romania, it ensures that it also complies with the local rules applicable to the automated evaluation of candidates.
5. The features' limits
5.1 Indicative results. Scores, verdicts and suggestions are estimates based on the available text. They may be incomplete or wrong, especially when the CV is short, in an unusual format or in a rarely encountered language. The confidence level shown must be taken into account.
5.2 No guarantee of outcome. Partnerd does not guarantee that a high score means a suitable candidate or that a low score means an unsuitable one. Hiring decisions and their consequences belong to the Client.
5.3 Changes. Partnerd may update the models and instructions used. Each change is versioned and visible in the log. Changes that significantly alter the behaviour of an evaluation feature are announced to the Client.
6. Liability
The parties' liability for the artificial intelligence features is subject to the limitations in the Terms of Use. The Client is liable for decisions made on the basis of the results and for compliance with its deployer obligations under the AI Regulation. Partnerd is liable for its provider obligations.
7. Updates
This document is updated as the AI Regulation comes into application and as the platform evolves, following the amendment procedure in the Terms of Use.
Contact for questions about the artificial intelligence features: hello@partnerd.ro