CodeSignal

Company overview

Founded in 2015 (originally as CodeFights, rebranded to CodeSignal in 2018) and headquartered in San Francisco, CodeSignal is a developer-focused skills platform built around technical assessments, structured coding interviews, and a learning product. Its hiring suite is anchored by Certified Assessments — role-based evaluations written by subject-matter experts and validated by in-house Industrial-Organizational (I-O) psychologists — layered with newer AI capabilities: an AI Interviewer that conducts and scores interviews, AI-assisted and "agentic" coding assessments (Cosmo, Claude Code, Cursor), and a blended AI-plus-human proctoring system that produces a "Suspicion Score." The company sells primarily to enterprise engineering and talent-acquisition teams and operates a Talent Science team that publishes blog-level guidance on fairness and adverse impact.

Regulatory exposure

As an automated/AI-assisted screening and interview tool used in employment, CodeSignal sits squarely within the scope of NYC Local Law 144, the EU AI Act's high-risk Annex III employment category, and Illinois/Colorado AI-hiring rules. Notably, CodeSignal's own published position is that its assessments are "unlikely to be classified as an AEDT" and that commissioning bias audits is the employer's obligation, not the vendor's — it offers to supply data for audits clients run but publishes no independent bias audit, names no auditor, and links no downloadable audit results. Its I-O-psychologist adverse-impact work supports EEOC/OFCCP framing but is not published as a transparent, dated audit. No EU AI Act FRIA template or deployer Article 27 guidance was found; AI-specific governance (ISO 42001, model cards, explainability statements) is absent, with security trust resources (SOC 2, ISO 27001, GDPR, Data Privacy Framework) gated behind a Trust Center.

Path to a higher score

CodeSignal could materially raise its score by commissioning and publicly posting a dated, independent NYC LL 144-style bias audit (e.g., BABL AI, Warden AI, Holistic AI) with a downloadable summary and the auditor named, and by repeating it annually. Publishing its adverse-impact / validity methodology as a citable whitepaper rather than blog prose, adding an AI system/model card and explainability statement for the AI Interviewer and Suspicion Score, releasing EU AI Act Article 27 deployer guidance plus a FRIA template, and pursuing ISO 42001 would address the AI-governance gaps. Finally, an explicit public statement on whether candidate data trains its AI models, a published sub-processor list, and a security/vulnerability disclosure contact would strengthen data-governance and post-market-monitoring transparency.