AI Transparency & the EU AI Act

Spotlite uses artificial intelligence to help candidates show their work and recruiters evaluate applications. The European Artificial Intelligence Regulation (AI Act, Regulation EU 2024/1689) strictly governs these uses — especially in employment, which is classified as "high-risk". This page describes, system by system, what our AI does, what it never does, your rights, and the obligations of our professional clients. It complements our AI & Data Charter.

1. Who is responsible for what#

The AI Act distinguishes two roles. Spotlite is the "provider" of the AI systems described here: we design them, document them and answer for their compliance. Recruiters and companies using our tools or our API are "deployers": they use them within their own hiring processes and have their own obligations (see section 6). The underlying foundation models (notably Gemini) are supplied by Google and run in European regions.

2. Inventory of our AI systems#

High-risk systems (Annex III, point 4 — employment and workers management):

• Candidate ↔ job match score: compares a CV and/or a video against a job offer and produces an indicative score (0-100), strengths and gaps.
• Video application analysis: extracts demonstrated skills, expertise level and an indicative employability score from a video.
• AI-led oral interview: our assistant "Lya" conducts an audio interview of 15 minutes maximum; a written synthesis (strengths, points to clarify, indicative score) is produced for the recruiter. Raw audio is never kept.
• Personality tests (MBTI-inspired, Big Five, DISC, Enneagram): questionnaires with deterministic scoring, plus an optional AI narrative analysis. They are indicative and not certified.
• Passive personality assessment: on request and from texts you provide (career journal), the AI infers an indicative personality profile, together with the reasoning that explains it.

Systems subject to transparency (Article 50):

• Dalia, our conversational career coach, and the recruiter Copilot: you are always informed that you are talking to an AI.
• Real-time translation in video calls, automatic subtitles and transcriptions.
• Blog articles written with the help of AI: they are systematically reviewed, edited and approved by our editorial team, which assumes editorial responsibility for them. As such, they fall under the Article 50(4) exemption and do not carry a mandatory AI label (commercial content under human editorial control, not the informing of the public on matters of public interest).

Minimal-risk systems: automatic content moderation (with human safeguards, see section 4), job-offer extraction from a URL, writing aids.

3. What our AI never does#

• No emotion analysis of candidates. Article 5(1)(f) of the AI Act prohibits emotion inference in the workplace context. Our application analyses and AI interviews assess neither emotions, nor stress, nor a person's "state of mind": only demonstrated skills and the professional content of what is said.
• No automated decisions. No Spotlite system automatically rejects, shortlists or sanctions a candidate. Every AI result is a decision aid that requires human validation.
• No social scoring, no biometric identification, no profiling based on origin, gender, religion, physical appearance or any other personal attribute: our internal model instructions explicitly forbid it.
• No misuse: our API terms prohibit using our systems for emotion inference, social scoring or any practice prohibited by Article 5.

4. Human oversight (Article 14)#

• Every score is explicitly indicative and every response of our evaluation API contains a non-removable aiNotice field: "AI-generated result, may contain errors, must never be the sole basis for a decision — human review is mandatory".
• AI interview syntheses shown to recruiters carry the same warning.
• In moderation, an account is never banned automatically: above a threshold, the case is flagged to a human administrator who decides.
• Recruiters keep access to the source material (full transcript, video, CV) to verify any AI result.

5. Your rights as a candidate#

• Being informed: when your application is analysed by AI or an interview is conducted by an AI, you are informed beforehand (explicit consent for the oral interview).
• Understanding (Article 86 — right to explanation): after an AI interview, the points the AI flagged as needing clarification are shared with you, and you can attach a short explanation that will be visible to the recruiter next to the synthesis. Personality assessments include the reasoning that led to the results.
• Contesting and obtaining human intervention (Article 86 and GDPR Art. 22): you can contest any evaluation, express your point of view and require that a human re-examines your file — with the recruiter concerned, or with Spotlite at contact-us@spotlite-app.com.
• Your data: raw audio of AI interviews is never kept (only the transcript and synthesis are); your access, rectification and erasure rights are described in our Privacy Policy.

6. Obligations of our professional clients (deployers)#

If you use the Spotlite Talent ATS or our B2B API, you are the deployer of a high-risk system and you commit to:

1. Displaying the aiNotice field as-is to anyone viewing an AI evaluation result — it may not be removed or rephrased.
2. Informing every candidate that their application is analysed with the help of AI (Article 26(7)). The aiNotice.candidateNotice field of every response contains ready-to-use wording in three languages.
3. Guaranteeing human review: never reject, filter or definitively rank a candidate solely on the basis of an AI score or recommendation.
4. Offering candidates a channel for explanation and contestation (Article 86).
5. Keeping logs of your evaluation calls for at least 6 months (Article 26(6)).
6. Not diverting the API towards prohibited uses (emotion inference, social scoring, biometric identification…).

Per-endpoint technical documentation is available in the API documentation.

7. Accuracy, limits and data#

AI outputs may contain errors, approximations or "hallucinations", even with correct input data. Scores are statistical estimates, not measurements: two analyses of the same file may vary slightly. This is precisely why human review is mandatory. Our AI processing runs in European hosting regions, the technical logs required for traceability are retained, and our contracts prohibit our providers from reusing your data to train their own models (see the AI & Data Charter).

8. Labelling of AI-generated content (Article 50)#

Conversational agents announce themselves as AI; the oral interview starts with an explicit consent text; AI-generated syntheses, analyses and narratives are flagged as such in interfaces and API responses. AI-assisted blog articles undergo human editorial review with assumed editorial responsibility: they benefit from the labelling exemption provided for in Article 50(4).

9. Compliance file — downloadable documents#

Our complete AI Act compliance file (version 1.0 of 6 July 2026) is public, viewable and downloadable from the Compliance file page:

• DOC-01 — Register of AI systems: inventory and classification of all our systems;
• DOC-02 — Consolidated technical documentation (Art. 11, Annex IV): architecture, models, data, performance;
• DOC-03 — Risk management system (Art. 9): risk register and mitigation measures;
• DOC-04 — Instructions for use for deployers (Art. 13 & 14);
• DOC-05 — Candidate information templates (Art. 26(7) & 86), in three languages;
• DOC-06 — Post-market monitoring plan (Art. 72);
• DOC-07 — Serious incident management procedure (Art. 73);
• DOC-08 — EU declaration of conformity (draft) (Art. 47, Annex V);
• DOC-09 — EU database registration file (Art. 49, Annex VIII).

Each document can be printed to PDF from the browser. The official documents are written in French.

10. Incidents and contact#

If you observe a serious malfunction of one of our AI systems (discriminatory result, error with consequences, circumvention of safeguards), report it to contact-us@spotlite-app.com — subject "AI Incident". We investigate every report and notify the competent authorities where the regulation requires it. For any question about this page: same address.