Built to be
audited.
AI in hiring is the part buyers ask about on the first call. Here is exactly what we do, exactly how the AI is constrained, and, just as importantly, what we will not claim.
- PASSWORD_HASH
- bcrypt
- TOKEN_AT_REST
- AES-256-GCM
- ACCESS_TOKEN
- short-lived + rotation
- FILE_ACCESS
- authenticated proxy
- TENANCY
- isolated at data layer
- AI_DECISIONS
- human-submitted only
Three commitments, in order of importance.
Emerging regulation treats solely-automated hiring decisions as a category of their own. Our architecture keeps a named human accountable at every consequential step, not out of caution, but because it is the thing that makes the AI usable at all.
A human decides
AI produces drafts. A named person reviews, can override any field, and only their submission is final. That holds for scorecards, for bias flags and for captured candidates alike.
Every AI claim is traceable
Scorecard bullets link to transcript timestamps with confidence values. Bias flags carry the exact text and its surrounding context. Nothing is a black-box score.
Bias detection protects candidates
It exists to catch unfair interviewer behaviour. It flags what the company asked, never what the candidate is.
What is in place today.
Every item below is implemented in the platform right now. Nothing on this list is roadmap.
- Multi-tenant data isolation
- Candidate and job records are scoped to the owning company or agency at the data layer.
- Role-based access control
- With department and location scoping, so access narrows to the part of the org a person actually works in.
- Rate limiting
- On authentication and API surfaces, with a dedicated origin guard for extension traffic.
- Passwords hashed with bcrypt
- Never stored in plaintext, never recoverable, only resettable.
- Short-lived access tokens
- With refresh-token rotation, so a leaked token has a narrow window.
- Integration tokens encrypted at rest
- Google, Microsoft and Zoom credentials are encrypted with AES-256-GCM.
- Files are never publicly readable
- Documents and resumes are served through an authenticated proxy, not public bucket URLs.
- Upload hardening
- Executables and macro-enabled Office documents are rejected; attachment size caps are enforced.
- Duplicate prevention
- Deterministic hashing computed identically on client and server.
- Audit trails
- Event logs, status history, job-order logs and email send logs.
- Captured candidates require human promotion
- Nothing enters the candidate database automatically.
- Named human sign-off
- Scorecards are drafts until a person submits them under their own name.
The vendors overstating this are the reason you have to ask.
Recruitment AI has a credibility problem, and it was earned. Here is where we stop short, stated plainly rather than buried.
We do not hold a certification we have not earned
We are not claiming SOC 2, ISO 27001 or "GDPR compliant" on this page. If you need our current certification status and timeline for a procurement review, ask us directly and you will get a straight answer.
We do not say all data is encrypted at rest
Integration tokens are encrypted with AES-256-GCM. Files are access-controlled and tenant-isolated. Universal field-level encryption is an active remediation plan, not a shipped feature, and we are not going to describe it as one.
We do not claim to eliminate bias
We detect and flag bias in interviewer behaviour and route it to a human. No vendor can eliminate bias, and any product telling you otherwise is describing something it cannot do.
Our AI does not screen anyone out
It ranks and surfaces. Nothing auto-rejects. Claiming otherwise would describe an automated decision with legal consequences we deliberately do not make.
We do not promise automated data deletion yet
There is no automated retention or purge regime today, and subject-rights requests are handled manually rather than self-service. Both are known gaps.
Bring us your security questionnaire.
We would rather answer it properly than have you discover the gaps in month three. Start a trial, or talk to us first.