Alva Labs

Dedicated Safe AI page states Alva is 'among the first companies in the world to be certified according to ISO/IEC 42001', names DNV as the certifying body, links a DNV certificate checker entry, and scopes certification across the proprietary psychometric tests, candidate platform AI layer, customer platform and internal AI governance systems.

https://www.alvalabs.io/trust/safe-ai

Help article documents an AI Management System (AIMS) built on ISO/IEC 42001 and the EU AI Act, enumerates the seven EU AI Act trustworthy-AI values Alva has adopted, and records that Unidot audited the AIMS implementation in February 2025 ahead of DNV certification.

https://help.alvalabs.io/en/articles/10654836-safe-ai-at-alva

'How AI Is Used in Your Hiring Processes' (dated 28 February 2025) gives feature-level explainability, describing LLM use for job-title cleaning and taxonomy matching, interview-question generation from personality traits, and automatic coding-test scoring against a 0-2 scorecard.

https://help.alvalabs.io/en/articles/10681646

'AI in Alva's Psychometric Tests' documents the modelling stack - machine learning for response-pattern trends, Item Response Theory for item effectiveness, and Bayesian inference via NumPyro for difficulty, validity and reliability estimates.

https://help.alvalabs.io/en/articles/10513028

Public help-center page offering the Alva Logic Test technical manual as a downloadable PDF, last updated 26 February 2025.

https://help.alvalabs.io/en/articles/4510873-technical-manual

No Annex IV technical documentation pack, model card, or system card was found across alvalabs.io/trust/safe-ai, alvalabs.io/trust/science, the help center, or web searches for 'Alva Labs model card / system card / instructions for use'.

Have a URL that disputes this?

Personality test product page states the test is 'DNV-certified' and 'Certified to ISO 10667-2', describing this as 'Independent third-party validation of scientific rigour and practical utility'.

https://www.alvalabs.io/hiring-system/assessments/personality-test

Science page states Differential Item Functioning (DIF) analysis is performed on all items before deployment with biased items flagged, revised or removed, and cites calibration on 1.5 million candidates with full item-bank recalibration every 18 months.

https://www.alvalabs.io/trust/science

Product page asserts 'Personality assessments show minimal group differences across gender, ethnicity, and age' but publishes no impact ratios, effect sizes, or demographic sample sizes to support it.

https://www.alvalabs.io/hiring-system/assessments/personality-test

Publicly downloadable PDF confirmed as 'Alva's Logic Test - Technical Manual.pdf', linked from the help center; the PDF body could not be parsed by the fetch tool, so its adverse-impact contents could not be independently verified.

https://drive.google.com/file/d/1rtSwbtF-tPJ0t0rTYJf6BLhXQbMvEgpY/view

No NYC Local Law 144 bias audit summary and no independent algorithmic bias audit by a named auditor (BABL AI, Warden AI, Holistic AI, ORCAA, DCI, Credo AI) was found on alvalabs.io/trust/safe-ai, the security page, the science page, the help center, or via targeted searches pairing 'Alva Labs' with those auditors and with 'bias audit' / 'adverse impact' / 'Local Law 144'.

Have a URL that disputes this?

States Alva's AI Task Force 'conducts risk and impact assessments on our AI systems' to identify how they could negatively impact individuals and communities and to explore mitigations - a provider-side impact assessment, not a deployer FRIA.

https://help.alvalabs.io/en/articles/10654836-safe-ai-at-alva

Assigns customers a concrete obligation to review and, if needed, correct AI feature outputs for accuracy, relevance and fairness, and provides ai@alvalabs.io as the channel for reporting AI-related harm - partial deployer guidance.

https://help.alvalabs.io/en/articles/10654836-safe-ai-at-alva

No Article 27 Fundamental Rights Impact Assessment template, checklist, or deployer-obligation guidance was found on alvalabs.io/trust/safe-ai, alvalabs.io/trust/security-gdpr, alvalabs.io/trust/science, the help center, or via searches pairing 'Alva Labs' with 'Fundamental Rights Impact Assessment', 'Article 27' and 'deployer'.

Have a URL that disputes this?

States candidate data stays in the EU and 'is never used to train third-party models'- an explicit training-data exclusion - alongside ISO/IEC 27001 and GDPR compliance.

https://www.alvalabs.io/trust/safe-ai

Security page documents AES-256 encryption at rest, TLS 1.2+ in transit, third-party key management via Thales with Key Access Justifications, annual external penetration testing, and a 24-month default data retention window, and links an ISO 27001 certificate and a security practices PDF.

https://www.alvalabs.io/trust/security-gdpr

Sets out the data categories collected, distinguishes Alva's three GDPR roles (processor, joint controller, independent controller) with the legal basis for each, and confirms processing is governed by a Data Processing Agreement.

https://help.alvalabs.io/en/articles/14594840-how-alva-labs-protects-and-processes-your-data

Documents a 24-month default retention clock from candidate creation with admin-configurable, country-specific retention periods and anonymisation of names, emails, test results and interview notes at expiry.

https://help.alvalabs.io/en/articles/8989961-data-retention

The sub-processor list linked from the security page (app.alvalabs.io/data-processors) did not render any sub-processor names publicly when fetched, so the roster appears gated behind the application; no SOC 2 report was found anywhere on the site.

Have a URL that disputes this?

States 'It is 100% optional to use these AI features', lets users customise coding-review scorecards and regenerate interview questions if unsatisfactory, giving concrete per-feature opt-out and override controls.

https://help.alvalabs.io/en/articles/10681646

Frames AI as 'built to support, not replace, human decision-making', names Human Agency and Oversight as the first of its seven adopted values, and requires users to review and correct AI outputs before relying on them.

https://help.alvalabs.io/en/articles/10654836-safe-ai-at-alva

Confirms ML models surface item-level trends but that Alva's psychometricians retain authority to adjust or remove test questions, keeping a human in the loop on model changes.

https://help.alvalabs.io/en/articles/10513028

Candidate-facing page states 'You receive the full report immediately after completing the test. It's the same report the recruiter sees. No hidden scores.', giving candidates symmetric visibility into the output used to evaluate them.

https://www.alvalabs.io/candidates

No per-jurisdiction compliance toggles, override audit logs, or documented candidate right to request human review of an assessment outcome were found on the candidates page, the Safe AI pages, or the help center.

Have a URL that disputes this?

Live public status page monitoring the Alva App (5 components, 99.97% uptime), Integrations (9 components) and email delivery, with viewable incident history and subscribe-to-updates notifications.

https://status.alvalabs.io

Publishes ai@alvalabs.io as a dedicated channel for reporting AI features that may harm individuals, and describes ongoing supplier and tooling review by the compliance team.

https://help.alvalabs.io/en/articles/10654836-safe-ai-at-alva

States continuous monitoring of assessments with full item-bank recalibration every 18 months, giving a published cadence for model revalidation.

https://www.alvalabs.io/trust/science

Documents annual penetration testing by external security experts as a recurring post-deployment assurance activity.

https://www.alvalabs.io/trust/security-gdpr

No continuous fairness-monitoring dashboard, model-update changelog for assessment scoring, or published vulnerability disclosure policy was found on status.alvalabs.io, the security page, the Safe AI pages, or the help center.

Have a URL that disputes this?

Trust hub page consolidating certifications, encryption, retention and GDPR posture, and linking out to the ISO 27001 certificate, a security practices PDF, data retention docs and the sub-processor list.

https://www.alvalabs.io/trust/security-gdpr

Help-center page (last updated 13 November 2025) offering the personality test technical manual as downloadable PDFs in English, Swedish and Norwegian plus an example candidate report.

https://help.alvalabs.io/en/articles/4522026-technical-manual-for-alva-s-personality-test

Full public product page for the adaptive logic test explaining the IRT and Bayesian methodology, single calibrated scale, per-question timing and DNV certification.

https://www.alvalabs.io/hiring-system/assessments/logic-test

Dedicated candidate-facing page covering report access, data ownership, result portability across employers, and retake options.

https://www.alvalabs.io/candidates

Customer-facing data processing explainer confirming DPA-governed processing and setting out Alva's controller/processor roles.

https://help.alvalabs.io/en/articles/14594840-how-alva-labs-protects-and-processes-your-data

No NYC Local Law 144 guidance, Illinois HB 2557 / AI Video Interview Act guidance, Colorado SB 24-205 guidance, or EU AI Act deployer checklist was found across the trust pages, help center, or targeted searches.

Have a URL that disputes this?