Ethics & Fairness

How you hire reflects what you value.

At Obscura, we believe every candidate deserves a fair, consistent, and evidence-based evaluation - regardless of background, identity, or circumstance. This page sets out our commitments in full. We stand by them.

Our Approach to Algorithmic Fairness

Obscura does not use protected characteristics as model inputs. Race, gender, age, disability status, and national origin play no role in our scoring algorithm. Candidates are assessed solely on their responses to standardised visual stimuli.

We conduct annual third-party bias audits through Perceptual Analytics Group. These audits assess score distributions across demographic groups and test for adverse impact at the construct level. Audit reports are available to all clients on request.

Our internal data science team reviews model performance quarterly. We take prompt action when performance variance across groups exceeds our internal tolerance thresholds.

Candidate Transparency

Candidates are informed before completing the assessment that their responses will be processed to generate a hiring signal for the recruiting organisation. We provide a plain-language candidate information sheet at the start of each assessment session.

Candidates may request access to their raw response data under applicable data protection law. Obscura Insight Reports are proprietary scoring outputs and are not candidate-facing documents; however, our Data Processing Agreement sets out the legal basis for this approach in full, and we recommend clients review it before deployment.

Independent Oversight

Our bias audit is conducted annually by Perceptual Analytics Group. Their findings are reported to our Board and summarised in our annual Transparency Report.

We are members of the Responsible Assessment Coalition and the AI in Hiring Working Group at the Institute for Employment Studies.

We have an Ethics Advisory Board comprising four independent members with backgrounds in organisational psychology, employment law, and computational social science. The Board meets quarterly and has standing access to our Technical Appendix.

Our Commitments

• We will not deploy Obscura in any context where it has not been validated against role-relevant performance criteria.
• We will publish an annual Transparency Report covering bias audit outcomes, adverse impact data, and model performance.
• We will provide clients with the information they need to use Obscura responsibly, including access to the Technical Appendix.
• We will take prompt action when evidence of bias or adverse impact is identified.
• We believe hiring should be human. Obscura is a tool to support that process, not to replace it.

A Note on Section 7

We are aware that several clients have raised questions about Section 7 of the Technical Appendix, which documents construct-level performance variance across demographic groups. We want to be direct: the data in Section 7 is important, and we encourage all clients to read it before deploying Obscura at scale.

Section 7 access requires a signed mutual NDA and completion of our Enterprise Onboarding Programme. This requirement exists because the data in Section 7 is technical and requires context to interpret accurately. We provide that context as part of onboarding.

We recommend reviewing Section 7 before deployment. We understand that most clients review it after. We are working to change that.

Questions

Obscura believes hiring should be human. We are here to support that process.