AIDisclosure

Effective 31 August 2026. Version 1.0. Published to meet the transparency duties in Regulation (EU) 2024/1689 (the AI Act), Articles 13 and 22 GDPR, and comparable US state law.

1. Where AI is used in this platform

Feature	What the AI does	Affects a person?
AI interview	Conducts a live spoken conversation, asks and follows up on questions	Yes
Transcription and translation	Converts speech to text, translates it	Yes
Interview analysis	Produces competency ratings, a written report and an overall score	Yes
Assessment grading	Grades free-text answers, awards partial credit, computes a percentage	Yes
Automatic pass or reject threshold	Moves an application to shortlisted or rejected when a configured score is crossed	Yes, and it can act without a human first
Best-fit and job-match reports	Ranks applicants against a role, or roles against a profile	Yes
Proctoring signals	Detects whether a face is in frame, and logs tab, focus and clipboard events	Yes, indirectly
Content generation	Drafts job descriptions, interview questions and assessment items for employers	No, an employer reviews it
Hiring assistant and voice helpers	Suggests next actions to the employer inside the dashboard	No

The underlying models are provided by Google (Gemini family) and are accessed through their API. Speech recognition may additionally use your browser's built-in speech service. Face-in-frame detection runs locally in your browser using MediaPipe and does not create a biometric template or identify you.

2. Legal classification, and who is responsible for what

Using AI to evaluate people for employment is classified as high-risk under Annex III(4) of the EU AI Act. Two different parties carry duties, and it matters which one you are dealing with.

• AIHire is the provider. We build the system. We are responsible for its design, its documentation, its logging, its accuracy statements, and for giving deployers what they need to oversee it.
• The employer is the deployer. They decide to use it, on which candidates, with which thresholds. They are responsible for human oversight, for informing candidates, for their own impact assessment, and for every hiring decision.

3. Known limitations and risks

We publish these because you should not be surprised by them, and because a deployer cannot exercise real oversight without knowing where the system is weak.

• Fabrication. Language models can state things confidently that are not true, including in a summary about a candidate. Read the transcript, not only the summary.
• Non-determinism. The same session analysed twice can produce slightly different wording and scores.
• Language and accent. Transcription accuracy is lower for strong accents, dialects, code-switching, speech differences and non-native speakers. A lower transcript quality can depress a score for reasons that have nothing to do with ability.
• Audio and network conditions. Poor microphones, background noise and unstable connections degrade both the interview and its analysis.
• Disability and accommodation. Proctoring signals such as gaze away from screen, face out of frame, or leaving the tab can be triggered by assistive technology, a screen reader, a medical condition, a caring responsibility, or a shared living space. They are not evidence of misconduct.
• Bias. Models inherit bias from their training data. We do not claim this system is free of bias in respect of any protected characteristic, and no vendor can honestly claim that. Deployers must monitor outcomes and, where their law requires it, commission an independent bias audit.
• Scope. The system evaluates what was said in a session against the criteria the employer set. It does not measure intelligence, honesty, personality, emotion, or fitness for a job in any general sense.
• Recording failure. A session can fail to record, upload or transcribe because of a device or connectivity fault. Absence of a recording means nothing about the candidate.

4. Human oversight: what employers must do

1. Assign a competent person to review AI Output before any adverse action, and give them authority to override it.
2. Read the transcript and evidence, not the score alone, before rejecting anyone.
3. Set automatic thresholds deliberately, review the outcomes they produce, and be prepared to justify them.
4. Tell candidates, before they are assessed, that an AI system is used and what it does. Some jurisdictions require specific wording and prior consent.
5. Offer a reasonable alternative to a candidate who cannot or does not wish to be assessed by AI, and offer accommodations on request.
6. Give candidates a route to request human review, and act on it.
7. Keep the records your law requires. Do not rely on this platform as your archive.

5. Your rights as a candidate

• To be told. You must be informed before an AI system evaluates you. This page and the notice on the pre-session screen serve that purpose.
• To an explanation. Where an automated result significantly affected you, you can ask the employer for a clear explanation of the role it played in the decision.
• To human review. You can require that a person, not a machine, reviews an outcome that has a legal or similarly significant effect on you, put your point of view, and contest it.
• To your data. You can request a copy of your recording, transcript, answers and scores, ask for corrections, and ask for erasure. See the Privacy Policy.
• To withdraw. You can end a session at any time and ask the employer for an alternative route.

Start with the employer, because they are the controller and the decision maker. If you cannot reach them, email privacy@aihire.io and we will route your request and tell you that we have.

6. Training, and what we do not do with your data

• We do not train models on your recordings, transcripts, answers, CV or scores.
• Our model providers are contractually barred from training on data sent through our API integration.
• We do not use candidate data to build a talent database, a benchmark, or a product we sell to anyone else.
• Product analytics use aggregated, de-identified counts only.

7. Accuracy, logging and documentation

We keep the technical documentation and event logging the AI Act expects of a provider, including records of session lifecycle, model configuration in force at the time, and the reason recorded for any automated stage change, so that an individual decision can be reconstructed and explained. Employers may request the information they reasonably need for their own impact assessment at privacy@aihire.io. We do not publish a single headline accuracy figure, because a figure produced on our own test set would not be a truthful predictor of accuracy on your roles, your candidates and your languages.

8. Reporting a problem with an AI result

If you believe an AI result was wrong, unfair, or produced a discriminatory outcome, tell us at ai-concerns@aihire.io. We investigate provider-side issues, and we will inform the affected employer. Serious incidents are handled under our incident process and, where the AI Act requires it, reported to the competent authority.

This page forms part of our Terms of Service.