Snapshot 12339
Normalized text
Scripts and page chrome removed; this is what change detection compares.
Responsible AI at Eightfold: evidence, not promises Responsible AI Trust is an engineering requirement. Talent decisions impact careers. Fairness in talent intelligence requires validation, not just intent. Eightfold AI trains its matching model on more than one billion anonymized career profiles. We test for bias on each model update, submit to independent audits annually, hold to strict data governance, and publish the documentation. How the Responsible AI program works: the framework, the measurements, the audits, and the humans who stay in charge. Explore Security at Eightfold ISO/IEC 42001 AI management FedRAMP Moderate authorized SOC 2 Type II and SOC 3 ISO 27001, 27701, 27017 Independent bias audits published under NYC Local Law 144 TX-RAMP certified IN SHORT Eightfold AI is the AI Native Talent Intelligence company. Its responsible AI is independently audited, certified to ISO/IEC 42001, and designed so people make the employment decisions. Prior to launch, Eightfold tests for bias on each model update and makes its model outcomes available to customers for review. The AI surfaces and explains; the hiring decision stays with the individual. The framework Five commitments govern each model we ship. Our T.R.U.S.T. principles are not a poster on a wall. Each maps to engineering practices, review gates, and measurable checks that run before and after each release, overseen by our external AI Ethics Council and an internal, cross-functional Ethical AI Committee. T Transparent. No black boxes. Match scores come with plain-language reasoning, candidates can see the skills behind their fit, and AI Interviewer summaries link to transcript evidence. R Reliable. Purpose-built, deterministic models. The same candidate and the same role produce the same score on each run, so outputs are stable and reproducible. U Upholds fairness with bias testing. Bias mitigation is built into the data pipeline and the model design, then verified with impact-ratio analysis and perturbation testing on each model update. S Safe and certified governance. Supported by independent audits, ISO/IEC 42001 AI management certification, FedRAMP Moderate authorization, and a documented, inspectable AI governance program. T Together, with human in the loop. Built for decision support. Our AI provides insights to empower human hiring teams, who retain full authority over selection and advancement decisions. The software does not autonomously reject a candidate or advance anyone through a hiring stage on its own. Independent verification Audited by people who are independent third parties. Claims are easy. We put our models in front of independent auditors and publish the results, as required by New York City Local Law 144. Independent bias audits. Our matching models and AI Interviewer undergo third-party bias audits, with summary results published under NYC Local Law 144. Audit documentation is available to customers to support their own compliance obligations. Certified AI governance. Eightfold is certified to ISO/IEC 42001:2023, the international management-system standard for artificial intelligence, alongside ISO 27001, 27701, and 27017. Ethics oversight. An external AI Ethics Council and a cross-functional Ethical AI Committee of engineering, legal, compliance, security, and product leaders review how our AI is designed, deployed, and monitored. Also maintained: SOC 2 Type II, SOC 3, FedRAMP Moderate authorization, GDPR and CCPA alignment, and a published subprocessor list, all available in the Trust Portal. People stay in charge The AI recommends. People decide. Read the AI transparency checklist Humans make the decisions. The AI provides recommendations and explainability to support hiring teams; it does not autonomously reject a candidate or advance one through a hiring stage on its own. Candidates can see the reasoning. Through personalized career site, candidates can see how they match to a role, the skills that align, the gaps in their experience. All visible before they hit submit. Interview summaries carry evidence. Each AI Interviewer insight links back to the transcript and response that supports it. Scores are role-specific. A match score assesses a candidate’s fit for one job; it is not a score that follows a person around to other roles or companies. The most transparent application process. Before I hit submit, it told me what it thought about my fit for the role and highlighted the skills and experience that informed that score, as well as the gaps. Job candidate, unsolicited public feedback FAIRNESS, MEASURED We test each model update for bias. Two checks run on each update to our matching models: impact-ratio analysis across demographic groups, and perturbation testing that rewrites anonymized, real resumes to change a single sensitive attribute, then re-scores them. Data: Impact ratios, 2026 model update Selection-rate ratios across hundreds of thousands of scored candidates. Perturbation testing We tested our model across eight attributes: Sexuality Race Religion Gender Veteran status Career gap Disability Age Changing any one of them, on an anonymized, real resume, produced no statistically significant effect on the match score. These figures come from our 2026 bias-audit summaries, published under NYC Local Law 144, and the Responsible AI white paper, where the full methodology is documented. Bias mitigation is tested with release, not treated as a one-time launch claim. Purpose-built, not general-purpose Why general-purpose LLMs shouldn’t power talent intelligence. We benchmarked our purpose-built matching model, part of the AI Native Talent Intelligence Platform, against nine frontier LLMs on a balanced pool of 10,000 anonymized applicants, measuring quality, bias mitigation, cost, and consistency. Task-specific architecture wins on the axes that matter for high-stakes decisions. Higher Accuracy. Scores higher than the frontier LLM Models tested. Measured bias mitigation. Stays above the 0.80 impact-ratio bar on intersectional subgroups, where most frontier LLMs fell below it. Fast processing. Processes each candidate far faster: frontier LLMs take more time and up to 78 times more computing power. Consistent scoring: Returns the same score on identical runs, while LLM scores drift. Read how we designed the benchmark and what we learned in Responsible AI: How we teach AI to be fair. Regulatory readiness Designed for compliance, wherever you hire. AI regulations are evolving quickly. We track regulations, engineer our platform to enable your compliance with them, and give customers the documentation and product controls to implement in accordance with their own obligations. In effect since 2023 NYC Local Law 144 Annual independent bias audits with published summaries, plus product workflows to support required candidate notices. Phasing in through 2027 EU AI Act A cross-functional readiness program: Instructions for Use, transparency documentation, governance and risk management, and independent bias-audit reporting, reinforced by ISO/IEC 42001. In effect Jan 1, 2026 Illinois HB 3773 Continuous bias auditing, no zip-code proxies, and configurable disclosure options to support employer notice duties under the amended Human Rights Act. In effect since 2020 Illinois AI Video Interview Act Configurable notice and consent flows, transcript-linked explanations, and human review for AI-assisted video interviews. In effect Jan 1, 2026 Texas TRAIGA Fairness testing, monitoring, and governance documentation that support deployer obligations. Go deeper The receipts. AI Interviewer bias audit results Published third-party bias-audit summary for AI Interviewer. Read now Matching model bias audit results Published third-party bias-audit summary for the matching models. Read now Security and compliance Data protection, certifications, and regulatory posture. Learn more AI you can trust: our Responsible AI blueprint How we build and govern Responsible AI. Read now The AI transparency checklist How to disclose AI use and build candidate trust. Read now Eightfold AI Interviewer, explained What it is and how it evaluates candidates. Read now Trust Portal Certifications, security documentation, and subprocessors. Learn more Responsible AI white paper Fairness methodology, model evaluation, and governance in depth. Read now AI Interviewer The flagship talent agent this trust program supports. Learn more FAQ Common trust questions, answered. What makes Eightfold different on responsible AI? Evidence. We evaluate our models for bias on each model update, submit to independent audits, and makes its model outcomes available to customers for review. How does Eightfold stay transparent with candidates? Candidates using our personalized career site can see the reasoning before they submit: their fit for the role, the skills that align, and the gaps. Is the AI fair and unbiased? Bias mitigation is built into the data and the model, then verified on each update with impact-ratio analysis and perturbation testing. Summary results are published under NYC Local Law 144. Who audits the AI? Independent third-party auditors. Our matching models and AI Interviewer undergo third-party bias audits, with summaries published under NYC Local Law 144. Eightfold is also certified to ISO/IEC 42001:2023. Does the AI reject candidates? No. The AI recommends based on fit to the hiring team and explains its reasoning. The AI does not autonomously reject a candidate or advance one through a hiring stage. People make the hiring decisions. Is Eightfold FedRAMP authorized? Yes. Eightfold maintains FedRAMP Moderate authorization, alongside SOC 2 Type II, SOC 3, and ISO 27001, 27701, and 27017. Talent Intelligence, human led. Explore Security at Eightfold Read the white paper Share Popup Title Share this article