AI Hiring Tools and Colorado Law: The Complete Employer Compliance Guide

If you had to bet on the Colorado AG's first SB 24-205 enforcement action, AI hiring tools would be the smart money. Here's why: **High visibility:** Employment affects everyone. Biased AI hiring tools generate sympathetic victims and clear narratives for enforcement. **Documented precedent:** Amazon's hiring AI scandal, NYC Local Law 144 (bias auditing for hiring AI), EEOC AI guidance — there's already a national enforcement playbook. **Easy to prove:** Disparate impact in hiring is straightforward to measure. Selection rates by demographic group are binary (hired/not hired) and well-studied. **Widespread adoption:** According to SHRM, 83% of employers use some form of AI in hiring. Most have never audited it for bias. This isn't speculation — it's the intersection of high impact, easy enforcement, and broad adoption. Colorado employers using AI in hiring should treat SB 24-205 compliance as their #1 AI governance priority.

Related: SB 24-205 compliance guide · bias audit guide · algorithmic bias explained

**Stage 1: Sourcing** - AI-powered job ad targeting (who sees your listing) - Automated candidate sourcing from databases → *Risk: If ad targeting algorithms show listings to one demographic more than another, that's potentially discriminatory sourcing.* **Stage 2: Screening** - Resume parsing and keyword matching - AI-scored applicant ranking - Automated rejection of "unqualified" candidates → *Risk: Resume screening AI is where Amazon's hiring tool failed. Training data bias and keyword proxies produce discriminatory filtering.* **Stage 3: Assessment** - AI-scored skills tests - Video interview analysis (tone, word choice, facial expressions) - Game-based assessments with AI scoring → *Risk: Assessment AI often uses proxies for cultural fit that correlate with protected characteristics. Video analysis tools have documented accuracy gaps across demographics.* **Stage 4: Selection** - AI-generated candidate rankings for hiring managers - Predicted job performance scores - Automated shortlisting → *Risk: The final recommendation stage compounds any bias introduced in earlier stages.* Every stage where AI influences the hiring outcome is a "consequential decision" under SB 24-205.

**1. Inventory all hiring AI:** Document every tool that uses AI in your hiring process — including features within your ATS, assessment platforms, and job boards. **2. Conduct bias audits:** Test selection rates across all protected classes at every stage where AI is involved. Apply the four-fifths rule and statistical significance testing. **3. Pre-application disclosure:** Before candidates submit applications, disclose that AI will be used in the screening/selection process. **4. Adverse action notice:** When AI contributes to rejecting a candidate, notify them that AI was involved, what type of data was used, and how to contest the decision. **5. Human review option:** Provide candidates with the ability to request human review of AI-driven rejection decisions. **6. Publish your AI policy:** Your website must include a statement about how you use AI in hiring and how you manage bias risk. **7. Record retention:** Maintain all hiring AI audit results, disclosure records, and contest outcomes for 3+ years.

**Name-based bias:** AI trained on historical data may score candidates with non-Anglo names lower. Multiple studies confirm name bias in resume screening AI. **Education institution bias:** AI that weights "prestigious" universities inadvertently discriminates based on socioeconomic status and race. **Gap penalty:** AI that penalizes resume gaps disproportionately affects women (maternity leave), caregivers, and people with disabilities. **Keyword optimization bias:** Candidates who know how to "optimize" resumes for AI screening (a skill correlated with socioeconomic status and education) score higher, creating selection bias. **Video interview bias:** AI analyzing facial expressions, tone, and word choice has documented accuracy gaps across race, gender, and disability. Candidates with speech differences, neurodiverse communication styles, or non-standard accents may be systematically scored lower. **Feedback loop bias:** When AI is trained on data about "successful" past hires, it replicates the demographic profile of your existing workforce — even if that profile reflects historical hiring bias.

New York City's Local Law 144 has been in effect since July 2023, requiring bias audits for automated employment decision tools. It provides a preview of what SB 24-205 enforcement looks like: **What NYC requires:** Annual third-party bias audits for AI hiring tools, public posting of audit results, candidate notice 10 days before AI screening. **What we've learned:** - Most employers were initially non-compliant (no audits existed) - Bias was widespread when auditing began — particularly in selection rate disparities - The notice requirement caused the most confusion (timing, format, content) - Vendors initially resisted sharing bias audit data; they've since adapted **How SB 24-205 goes further:** - Broader scope (not just hiring — all consequential decisions) - Self-serve auditing is acceptable (no third-party requirement) - AG notification within 90 days for discrimination incidents - Affirmative defense available through NIST AI RMF alignment Colorado employers can learn from NYC's rollout: start auditing now, solve the notice workflow, and build your evidence trail before enforcement begins.

**Week 1:** - Inventory all AI used in hiring (including ATS features you may not realize are AI) - Identify which stages involve consequential decisions - Request bias testing data from all hiring AI vendors **Week 2:** - Run initial bias audits on your applicant data (selection rates by demographic group) - Identify the highest-risk AI systems (those with the most bias or the most applicants affected) **Week 3:** - Draft and publish your AI-in-hiring governance policy - Implement pre-application AI disclosure notices - Create adverse action notice templates with AI disclosure language **Week 4:** - Set up human review process for contested AI decisions - Configure ongoing monitoring (monthly automated checks recommended) - Generate first evidence bundle documenting your compliance program CO-AIMS automates steps 2-4. Most employers are fully compliant within 2-3 weeks of setup. Plans start at $199/month.