AI Transparency

Last updated: September 11, 2026

Electryion is built on AI models that draft, sort, score and suggest. This page explains what those models do, where their limits are, and which decisions always stay with a person. It also documents how we handle recruitment features, which fall under stricter rules in the EU AI Act.

1. The core rule: AI suggests, you decide

No AI employee in Electryion sends, publishes, hires, rejects or commits anything to a third party on your behalf without your action, with one explicit exception you switch on yourself (scheduled social publishing). Concretely:

• Email replies are drafted into a review queue. You read, edit and press send.
• Candidate outreach is prepared as text you copy or send yourself.
• Candidate scores are advice with reasoning attached. The system cannot reject or advance anyone by itself.
• Job postings, contracts and requirement rewrites are drafts until you publish them.

2. Recruitment AI (EU AI Act, Annex III)

AI used for recruitment and candidate selection is classified as high-risk under the EU AI Act. Electryion's recruiter includes CV screening and candidate matching, so we apply the following safeguards:

• Human oversight by design. Screening produces a score, the evidence behind it, and open questions. It never changes a candidate's status, never filters people out of your view, and never triggers an automated rejection. There is no setting that turns this off.
• Logged assessments. Every AI assessment of a candidate is recorded with the model used, the score, the reasoning and the job it was scored against. You can review this history per candidate, and it is retained in the workspace audit log.
• Job-relevant criteria only. Screening prompts instruct the model to judge skills, experience, certifications and availability, and to disregard age, gender, nationality, ethnicity, religion, health, marital status and photographs. Models are not perfect at this: that is precisely why a person reviews every result.
• Explainability. Scores come with quoted evidence from the CV rather than an opaque number, so you can check whether the reasoning holds.
• Your responsibility as employer. You remain the decision-maker and the data controller for your hiring process. Electryion provides decision support; it does not make employment decisions.

3. Candidate data and sourcing

Sourced candidate profiles come from third-party data providers listed on our Sub-Processors page. We do not scrape candidate accounts using your credentials, and we do not collect data from behind logins.

• Search results that you never act on are deleted automatically after the retention window (180 days by default).
• Deleting a candidate deletes their CV file as well.
• Candidates can request access to or deletion of their data via [email protected], and we will act on it for the workspaces involved.

4. Which models we use

Electryion uses large language models from OpenAI (see Sub-Processors) for generation, classification and scoring. Your content is not used to train those models. Cheaper models handle bulk classification; stronger models handle text you actually read or send.

5. Known limitations

We would rather state these plainly than let you discover them:

• Models can be confidently wrong. Draft replies mark facts we could not verify with a placeholder so you fill them in instead of trusting a guess.
• Scores are estimates, not measurements. Two similar CVs can score a few points apart for reasons that do not matter.
• Sourcing data comes from third parties and can be outdated or incomplete.
• Salary and market indications are model estimates, not verified market data.

6. Questions

For questions about how a specific feature decides something, or to request the audit record of an assessment, email [email protected].