Skip to main content
This page is the information JobMojito, as the provider of an AI system, gives to the companies that use it to hire, the deployers (EU AI Act, Articles 13 and 26). File it with your own AI Act records. It describes the system as it works today and marks every capability with its availability. It is not a conformity statement and not legal advice: your legal team decides how the Act applies to your use. Availability labels used below:
  • Available: live for every company.
  • Available, per company: live, switched on for a company by JobMojito on request.
  • Q4 2026, Q1 2027: on the roadmap for that quarter.

​
1. System description and intended purpose

JobMojito helps a hiring team run structured, comparable first-round assessments. It has four AI functions: Intended purpose. Decision support for a qualified person who reviews candidates for a job. The output is one input to that person’s decision, next to the CV, the interview recording and their own judgment. High-risk category. Use in recruitment and selection, including analysing applications and evaluating candidates, falls under Annex III, point 4(a), of the AI Act. Treat JobMojito as a high-risk AI system in your records. Not intended for:
  • Final hiring decisions without a person reviewing the result. See section 5 for the one setting that can reject automatically and how to keep it off.
  • Inferring emotions, personality, health, disability or other protected characteristics from a candidate’s face, voice or wording. JobMojito does not do this and must not be configured or prompted to do it.
  • Building biometric identification datasets from interview recordings.
  • Decisions about current employees (promotion, termination, monitoring).

​
2. Roles and the deployer checklist

Under GDPR, JobMojito is your processor and you are the controller (see Responsible use). Deployer checklist (Article 26):
  1. Name the people who oversee the system and make sure they have the competence and authority to override it: recruiters and hiring managers who review results (section 5).
  2. Use it as intended. Keep a person in every decision; set pre-screening rules to “Mark for review” if you do not want automatic rejections (section 5).
  3. Control your input data. The questions, rubric and pre-screening rules you configure must be relevant to the job and must not be proxies for protected characteristics.
  4. Inform candidates that AI is used and how (section 6). Inform workers’ representatives before use where your national law requires it.
  5. Monitor the system in use: review outcomes per position regularly (section 8) and tell JobMojito about any risk or serious incident you notice (section 11).
  6. Keep the logs under your control for at least six months, or longer where your law requires it (section 7).
  7. Carry out a data protection impact assessment using this page and the documents in the TrustCenter. Public bodies and certain other deployers also need a fundamental rights impact assessment (Article 27).
  8. Answer candidates who ask for an explanation of a decision in which the system’s output played a part (Article 86): the result page shows the scores with their evidence (section 5).

​
3. Models, subprocessors and hosting

The current list of subprocessors, with the purpose, location and transfer basis of each, is in the JobMojito TrustCenter. The TrustCenter also holds the architecture and security documents and the Data Processing Agreement. Models change as providers release new versions. Material changes are announced as described in section 10. JobMojito does not train AI models on your candidates’ data; the models are used through their providers’ APIs under data processing agreements. Hosting. Your company’s data is stored in the EU (Ireland: database and processing in AWS eu-west-1). Some AI providers process requests outside the EU/EEA; those transfers rely on the EU Standard Contractual Clauses or the EU-US Data Privacy Framework, as listed per provider in the TrustCenter. Enterprise customers can have a dedicated database project of their own.

​
4. Inputs, outputs and their limits

Inputs: the candidate’s spoken or written answers, the audio and video of the interview, the CV and form answers they submit, and the role description, questions and rubric your team configures. Outputs: transcripts, scores (0 to 10) with explanations and quoted evidence, rubric verdicts per expectation (met, partly met, not discussed), pre-screening results, integrity flags and reports. Every output is shown to your staff in the admin and can be exported. Known limits:
  • Languages. The interview interface is available in 11 languages; speech recognition covers more. Scoring quality is measured per language (section 9) and is lower for languages with less test data. Review results in less common languages with extra care.
  • Transcripts. Speech recognition makes errors, more often with strong accents, poor microphones or background noise. Scores are based on the transcript; the recording is the reference when in doubt.
  • Pronunciation scores measure articulation compared with a reference speaker model. They do not measure whether a listener understands the candidate, and they are lower for many accents that are perfectly intelligible. Do not use them as a language-proficiency test.
  • Transcript sentiment rates the tone of the wording of the answers (text only, not voice or face). It is part of the answer score by default; set its weight to 0 in the position’s scoring settings if you do not want it to count.
  • Reading detection (candidate reading answers from a screen) was calibrated to a precision of 0.91 and a false-positive rate of 0.06 on labelled interviews. About 6 in 100 candidates who do not read can still be flagged.
  • Voice check. A voiceprint is taken from the microphone test at the start and each answer is compared with it. Shared rooms, headsets and poor audio cause false “different voice” or “multiple voices” flags.
  • Scores are relative to the rubric your team set. A score is not a measure of a person’s general ability or character.

​
5. Human oversight

What reviewers see. For every candidate: the recording, the transcript, each answer’s score with the AI’s reasoning, the rubric verdicts with quotes, a breakdown of how the overall score was calculated, the integrity flags with their evidence, and a notice that the content is AI-generated. How they override it:
  • Decisions (shortlist, move to the next step, reject, select) are made by a person in the admin. Each decision records who made it and when. With the audit log (Available, per company) every decision, view and change is also written to an append-only log with the user and the reason.
  • Integrity flags can be added or removed by a recruiter; manual changes are marked as such.
  • Scores can be adjusted with a reviewer’s comment and score adjustment, which is shown next to the AI score.
  • Rubric and questions are edited by your team; the AI does not change them.
Automatic decisions: pre-screening rules only. The AI interview never accepts or rejects a candidate on its own. Pre-screening can: a rule whose “When Unfulfilled” setting is Reject, or a CV match below the position’s Reject threshold, rejects the application automatically, and an application that passes all rules can move on automatically. To keep a person in every pre-screening decision:
  1. In the position’s pre-screening settings, set When Unfulfilled to Mark for review for every rule.
  2. Leave the CV match Reject threshold empty (no automatic rejection) and the Accept threshold empty if you also want a person to approve every pass.
Results then wait in Pending decisions for a reviewer. We recommend this configuration for deployers in the EU.

​
6. Transparency to candidates

Available:
  • Before the interview the candidate sees a notice that they are interviewed by an AI, why, what happens to their answers (recorded, summarised and scored against the role’s criteria, then reviewed by a recruiter), how their data is handled and their rights.
  • On their result (when you share feedback) the candidate sees that the score and feedback are AI-generated, that they are not a psychological or personality test, and that a person reviews them.
  • Talk to a person instead. With “Allow a human interview instead” switched on for an interview or a position step, the candidate can ask for an interview with a person, with a reason. The request reaches your recruiters, who propose times or decline. Switch it on wherever you want to offer an alternative to the AI interview.
Available, per company: application consent records (what the candidate agreed to, which privacy notice and when) and a company-specific privacy notice. Your own job adverts, invitations and careers pages should also say that AI is used. Reusable wording is in Responsible use.

​
7. Data governance and logging

Stored per candidate: profile and contact details, applications, form answers and files, recordings, transcripts, AI analyses, scores, flags, decisions and the e-mails sent. Recordings and files are in private storage reachable only through signed links. Logs: To keep logs for six months or longer, export decisions and results to your own systems on a schedule (CSV export, API or webhooks), or agree a retention period with JobMojito. Retention and deletion. Data is kept while your account is active unless your DPA sets another period. Candidates can delete their account; recruiters can delete candidates; a candidate’s own data export is produced on request. Details: Data export, retention and deletion. Configurable retention periods per data type: Q1 2027.

​
8. Bias and fairness monitoring

Built in (Available):
  • The same questions, rubric and scoring for every candidate of a position.
  • Scoring works on the transcript of the answers and the position’s rubric. The prompts tell the models not to penalise speech-to-text errors, filler words or imperfect phrasing, so the way a candidate speaks does not lower the content score.
  • Evidence quotes are checked against the transcript; a quote that the candidate did not say is discarded.
Fairness dashboard (Q4 2026): pass rates and average scores per group, with adverse-impact ratios. What a deployer should review each month until then, from the results export:
  • Pass and rejection rates per position and step, and whether any group of candidates is passed or rejected noticeably less often (for example by country or interview language).
  • Results where reviewers overrode the AI score or removed flags, and why.
  • Integrity flags per position, to spot settings or environments that produce many false flags.
  • Pre-screening rules: whether each rule is still necessary for the job.

​
9. Accuracy, robustness and cybersecurity

  • Scoring quality is measured with a benchmark that replays real, anonymised interview answers through the production scoring prompts on each model, per language, and compares the results with a reference judge, including checks that quoted evidence is real.
  • Interview robustness is tested end to end with a scripted candidate talking to the live interviewer (latency, turn-taking, interruptions) before releases.
  • Fallbacks. Every AI step has a chain of providers; if one fails or is slow, the next one answers.
  • Database and API changes run through automated test suites against a production-like copy of the database, and through a staging environment.
  • Security. TLS 1.2 or higher in transit, AES-256 at rest, row-level access control per company, signed links for files. JobMojito is working toward SOC 2 and ISO 27001. The security and architecture documents are available in the TrustCenter.

​
10. Change management

  • Changes ship continuously and without downtime; they are listed in the changelog.
  • New features that change what candidates experience or what you pay for ship switched off and are enabled per company, interview or position.
  • Changes to AI models and prompts apply to all companies; they are tested on the benchmark in section 9 before release and listed in the changelog.
  • A separate test company (sub-company) can be set up to try configurations before candidates see them. See Releases and test environments.

​
11. Contacts and incident reporting

  • Questions about this page, the AI functions or a candidate’s request for an explanation: support@jobmojito.com.
  • Security issues: security@jobmojito.com.
  • Serious incidents and risks. If you see the system cause or contribute to harm, a malfunction affecting decisions, or a risk to candidates’ rights, tell us at support@jobmojito.com without delay with the company, position and result concerned. We investigate, inform you of the outcome and report serious incidents to the market surveillance authority as the Act requires of a provider. Your own reporting duties as a deployer remain.