Exam Center
Book a demo
Security

Responsible AI — four commitments.

AI is how some of this works. It is not why you would buy it, and it never gets the last word.

Where AI runs across an exam, and what it never doesAcross the life of one exam: at authoring, translations are drafted with AI. At publish they are frozen. During the live exam, detectors run on the platform and no AI provider is called at all — which is why a provider outage or rate limit cannot reach a running exam. After submission, answers are evaluated. Every AI output is recorded as AI-produced, re-evaluation is audited, and a person can override anything. What it is never asked to do: summarise a session for a reviewer, match a face against an ID photo, or decide an outcome.THE LIFE OF ONE EXAMBEFORE PUBLISHAuthoringtranslations draftedAI PROVIDER CALLEDAT PUBLISHPublishtranslations frozenNO PROVIDER CALLLIVEThe examdetectors run on-platformNO PROVIDER CALLAFTER SUBMITSubmissionanswers evaluatedAI PROVIDER CALLEDAN OUTAGE OR A RATE LIMIT CANNOT REACH A RUNNING EXAMEvery output is attributable, and reversibleNOTHING AI PRODUCES IS FINALrecorded as AIthe score says which produced itre-evaluationaudited, every timeoverridea person can change anythingWhat it does not doUNBUILT, OR OUT OF SCOPEsummarise a session for a reviewermatch a face against an ID photodecide an outcome
01

AI proposes, a person decides

No AI output is a verdict. Detections become evidence for a human decision, and AI answer evaluation can be set to propose a score that a grader confirms.

AI proposes, a person decidesNo AI output is a verdict. Detections become evidence for a human decision, and AI scores can be set to propose rather than apply.AI OUTPUTA person decidesACCEPT OR REJECTRESULTNO AI VERDICT
02

AI never runs during an exam

Translation happens at authoring and is frozen at publish. Answer evaluation runs after submission. The exam runtime does not call an AI provider, so an outage or a rate limit cannot reach a live exam.

Where AI runs, and where it does notAI drafts translations at authoring and evaluates answers after submission. It does not run while a candidate is sitting the exam, so a provider outage cannot reach a live paper.AuthoringAI DRAFTSHUMAN OKAYSThe examNO AINO CALLSAfter submitAI DRAFTSHUMAN OKAYSTHE RUNTIME CALLS NO AISo an outage cannot reach a live examTRANSLATION FROZEN AT PUBLISHEVALUATION RUNS AFTER SUBMIT
03

Every output is attributable and reversible

AI-produced scores are recorded as such, re-evaluation is audited, and a human can override anything.

Attributable and reversibleAI-produced scores are recorded as such, re-evaluation is audited, and a human can override anything.Every AI output is recorded as oneScore proposed by AIMODELConfirmed by graderS. RAORe-evaluatedAUDITEDOverridden by graderS. RAOA HUMAN CAN OVERRIDE ANYTHING
04

You control the sensitivity

Detector thresholds, severities and risk weights are yours to set per account and per assessment. We do not decide how strict your exams are.

You control the sensitivityDetector thresholds, severities and risk weights are set per account and per assessment by the institution.Your settings, not oursPractice testSemester examCertificationTHE SAME PLATFORM, THREE STRICTNESSES
05

What AI does not do here

It does not summarise sessions for reviewers, does not match faces against ID photos, and does not decide outcomes. Those are either unbuilt or deliberately out of scope.

What AI does not do hereIt does not summarise sessions for reviewers, does not match faces against ID photos, and does not decide outcomes.Not builtREVIEW SUMMARIESFACE MATCHINGLIVENESS CHECKSBuiltTYPED DETECTIONTRANSLATION DRAFTSANSWER EVALUATIONTHE LEFT COLUMN IS A ROADMAP, NOT A FEATURE LIST
AI flags.Detectors produce typed evidence with a confidence score. Nothing is a verdict.
People decide.A reviewer accepts or dismisses each violation, then records one decision with a reason.
Candidates can appeal.Every decision is appealable, and the appeal is tracked to its own outcome.

Bring one paper. We'll run it live.

Thirty minutes, your own exam, no slide deck.