> ## Documentation Index
> Fetch the complete documentation index at: https://docs.exterview.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Responsible AI

> Responsible AI at Exterview, aligned to the widely used principles of fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability: mandatory human review, explainable scores, disclosure of AI interaction, and no autonomous adverse decisions.

## Purpose

How Exterview keeps AI-assisted evaluation fair, explainable, and under human control. Our approach aligns with the responsible-AI principles Microsoft and others use: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. This page states what we do under each; it describes Exterview's own practices and is not a claim of Microsoft certification.

## Our commitments

<CardGroup cols={2}>
  <Card title="Human review">
    People review and approve outcomes. AI assists; it does not decide.
  </Card>

  <Card title="No autonomous adverse decisions">
    No fully autonomous decision rejects or disadvantages a candidate.
  </Card>

  <Card title="Explainability">
    Every score traces to the evidence and rubric behind it.
  </Card>

  <Card title="Content-based scoring">
    Evaluation is based on what a candidate says, not face, tone, or biometrics.
  </Card>

  <Card title="Disclosure of AI interaction">
    Candidates are told when they are interacting with an AI system, and consent
    is captured.
  </Card>

  <Card title="Fairness by design">
    Rubrics are role-relevant and applied consistently, to reduce bias.
  </Card>

  <Card title="Reliability and safety">
    We are building an automated system to check AI outputs for quality and safety
    before and after they're used, before a person acts on them.
  </Card>

  <Card title="Privacy and security">
    Your data is kept separate from other customers' and is never used to train
    shared models.
  </Card>

  <Card title="Accessibility and inclusion">
    Evaluation surfaces are designed with accessibility in mind, and candidate
    accommodations are available on request.
  </Card>

  <Card title="Accountability">
    Named human owners are accountable for decisions, with a full audit trail.
  </Card>
</CardGroup>

<Warning>
  Exterview does not make autonomous hiring decisions. Reports are decision
  support; a human always owns the outcome.
</Warning>

<Note>
  If a customer turns on identity verification, it works by checking a photo ID
  against a face photo. That check is handled separately from scoring and is not
  used to make any hiring decision, it only confirms who the candidate is.
</Note>

## How this maps to responsible-AI principles

Exterview's practices line up with the six principles most enterprises, including Microsoft, use to evaluate responsible AI.

| Principle            | What Exterview does                                                                                                                                                                                            |
| -------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Fairness             | Role-relevant rubrics applied consistently, content-based scoring, human review, and traceability, not opaque trait inference.                                                                                 |
| Reliability & safety | We're building an automated critic layer to check AI outputs for quality; today this is a mix of automated and manual checks. See [AI Observability](/monitor/ai-observability).                               |
| Privacy & security   | Each customer's data is kept separate, encrypted, and never used to train shared models. See [Security](/trust/security), [Privacy & Residency](/trust/privacy-residency), and [Models We Use](/trust/models). |
| Inclusiveness        | Accessibility considered in the design of evaluation surfaces, with accommodations available on request.                                                                                                       |
| Transparency         | Explainable scores that trace to evidence, disclosure that a candidate is interacting with AI, and documented capabilities and limitations.                                                                    |
| Accountability       | Mandatory human review, named decision owners, no autonomous adverse decisions, and an audit trail. See [Auditability](/trust/auditability).                                                                   |

## Capabilities and limitations

Being transparent means being clear about the edges of the system, not only its strengths.

<CardGroup cols={2}>
  <Card title="What it is">
    Decision support: structured evidence and a score to help a person decide faster and more consistently.
  </Card>

  <Card title="What it is not">
    A decision-maker. It does not autonomously hire, reject, or rank a candidate
    out of the process.
  </Card>

  <Card title="Where it needs care">
    Below the calibration volume of a new deployment, treat scores as directional
    and keep human review especially close. See [Running a Fair
    Pilot](/best-practices/running-a-fair-pilot).
  </Card>

  <Card title="What it does not infer">
    It does not score on face, voice tone, emotion, or other biometrics.
  </Card>
</CardGroup>

## FAQs

<AccordionGroup>
  <Accordion title="Does the AI reject candidates on its own?">
    No. No fully autonomous adverse decision is made. A human reviews and decides.
  </Accordion>

  <Accordion title="Are candidates told they are interacting with AI?">
    Yes. Candidates are informed when a session is AI-conducted, and consent is
    captured before the session begins.
  </Accordion>

  <Accordion title="Does Exterview analyze faces, tone, or emotion?">
    No. Scoring is content-based, grounded in what a candidate actually says or
    submits. Separately, if a customer turns on identity verification, that
    feature checks a photo ID against a face photo purely to confirm identity, it
    plays no part in scoring or any hiring decision.
  </Accordion>

  <Accordion title="Can we explain a score to a candidate or auditor?">
    Yes. Scores trace to the underlying evidence and the rubric version applied.
  </Accordion>

  <Accordion title="How is bias managed?">
    Through role-relevant rubrics, consistent application, human review, and
    traceability, not opaque trait inference.
  </Accordion>

  <Accordion title="How do you keep the AI reliable in production?">
    A critic layer evaluates and scores each output. We're building automated
    observability over every step; this is being rolled out progressively. See [AI
    Observability](/monitor/ai-observability).
  </Accordion>

  <Accordion title="Is the platform accessible?">
    Evaluation surfaces are designed with accessibility in mind, and candidate accommodations are available on request. Contact your Exterview representative to arrange them.
  </Accordion>
</AccordionGroup>

## Related

<CardGroup cols={2}>
  <Card title="AI Observability" href="/monitor/ai-observability">
    How outputs are scored, monitored, and kept reliable.
  </Card>

  <Card title="Auditability" href="/trust/auditability">
    How every score traces to its evidence and reviewer.
  </Card>

  <Card title="Security" href="/trust/security">
    Access, encryption, and keeping each customer's data separate.
  </Card>

  <Card title="Compliance FAQs" href="/trust/compliance-faqs">
    Straight answers for security and procurement reviewers.
  </Card>
</CardGroup>