AI testing. Results included.
Flip Testing — ATS
Identical candidate profiles with demographic indicators changed to detect differential scoring or recommendations.
400 pairwise flip tests across 8 job types · 800 API callsPrompt Bias Testing
Identical prompts tested under different demographic profiles and scored by an independent judge for differential treatment.
959 paired cases · up to 9 demographic axes · independent judgePenetration Testing
Independent testing covering prompt injection, jailbreak attempts, data exfiltration and attempts to extract system instructions.
Annual third-party assessment · tracked remediation · Trust Centre evidenceRisk of Harm Detection
Evaluation of Luna's ability to recognise and safely respond to expressions of risk of harm to self.
100 vignettes · 5 severity tiers · 5-part clinical rubricMental Health Concern Recognition
Recognition of possible delusional ideation or psychotic features without reinforcing the belief.
200 vignettes · 2 agents · 5 concern categoriesSafeguarding Concern Recognition
Contextual safeguarding recognition and useful guidance without overstepping professional scope.
80 vignettes · 4 safeguarding domains · 5-part rubricEvidence, not a trust-us statement.
Tests use scripted or paired synthetic cases, predefined scoring rubrics and an independent judge or assessor. No real users are involved and production records are not read or written.
Results describe a defined test set—not a blanket claim that an AI system can never fail. Production monitoring, human oversight and change-triggered retesting remain part of the control.
Fairness review threshold · maximum 5% disparityQuestions deserve a named owner.
Ask our privacy team about the methods, safeguards or findings in our published evaluations.