01
Ingest all signals
Payroll, bank, ID, document uploads, criminal, sanctions, and prior verifications are normalized into a structured candidate record.
Ethur AI · Claude-powered review
An AI adjudicatorthat shows its work.
Ethur AI is a Claude-powered agent that reviews every candidate file the way your best adjudicator would — reading payroll, bank, and ID signals together, flagging contradictions, and explaining its reasoning in plain English.
Model SOC 2 isolationNo PII used for trainingHuman-in-the-loop fallbackFull reasoning audit log
AI adjudication
Claude · explainable · audit-logged
94%
Auto-clear accuracy
11x
Faster than human adjudicators
100%
Reasoning logged
0
Black-box decisions
/ The realityBlack-box AI
Black-box AI
doesn't pass legal review.
Most AI-driven screening tools spit out a score with no explanation — which is exactly what FCRA, EEOC, and state adjudication law won't accept. Ethur AI reasons in plain language, cites the data points it used, and routes edge cases to humans by design.
Without Ethur
- Opaque risk scores you can't explain
- Hallucinated 'red flags' on clean candidates
- No audit trail for adverse-action defense
- AI trained on candidate PII without consent
- One model trying to do everything badly
With Ethur
- Plain-English reasoning on every decision
- Cited source-of-record for every claim
- Full audit log per candidate, immutable
- Zero PII used for model training, ever
- Specialist agents per check type
/ Built for youWhat Ethur AI
What Ethur AI
actually does.
Document authenticity review
Inspects paystubs, W-2s, offer letters, IDs, and diplomas for AI generation, tampering, and forensic inconsistencies.
Cross-source reconciliation
Reads payroll, bank, and ID data side-by-side — flags name, employer, and income mismatches a human would miss in a 30-second review.
Adjudication assistant
Recommends clear / review / adverse for each check, with cited reasoning your team can accept, override, or appeal.
Adverse-action drafting
When required, drafts FCRA-compliant pre-adverse and adverse notices with the specific findings attached.
Pattern detection across candidates
Spots ring fraud, repeated AI templates, and synthetic-identity clusters that single-candidate review can't see.
Explainability API
Every AI decision is queryable: which data, which model, which prompt, which version. Built for legal and audit teams.
/ How it worksReasoning,
Reasoning,
not just scoring.
Every candidate file passes through specialist agents that read, reconcile, and recommend — with full traceability.
02
Specialist agents review
Document, identity, employment, and income agents each evaluate their domain and write reasoning to the audit log.
03
Reconciliation pass
A senior agent checks for cross-source contradictions and edge cases, then composes a unified recommendation.
04
Human routing
Anything below configured confidence thresholds routes to your adjudicators with the agent's reasoning attached as context.
Ready to see Ethur on your data?
Start a free trial in minutes, or get a tailored walkthrough with our team.
/ FAQ
Common questions
Which model do you use?
Anthropic Claude is the primary reasoning model, with task-specific models for forensic document analysis. We pin model versions per tenant for reproducibility.
Is candidate data used for training?
Never. Candidate PII is processed in isolated, zero-retention contexts and is never used to train any model — ours or our providers'.
Is this FCRA-compliant?
Yes. AI is positioned as a recommendation engine, not the decision-maker. Adjudication authority remains with your team, and every recommendation includes the underlying findings.
What about EEOC and adverse-impact testing?
Ethur AI is regularly tested for adverse impact across protected classes; reports are available to enterprise customers under NDA.
Can we override the AI?
Always. Every recommendation can be accepted, modified, or overridden, and the override is logged with the adjudicator's reason.