Colorado's AI Act Takes Effect: A Practical Audit of Your Hiring Tools

Key takeaways
- Enforcement date: The Colorado Artificial Intelligence Act (SB 24-205) becomes enforceable law on June 30, 2026 (ai2.work).
- Risk classification: Algorithms that significantly influence employment decisions are classified as high-risk AI systems (shrm.org).
- Compliance requirement: Deployers must perform impact assessments at least annually and within 90 days of any major modification (fisherphillips.com).
- Financial exposure: A single system affecting 1,000 Colorado applicants without proper disclosure could lead to up to $20 million in potential exposure (ai2.work).
As of June 30, 2026, the Colorado Artificial Intelligence Act (SB 24-205) is officially enforceable law (ai2.work). This legislation, passed by the Colorado legislature on May 8 (shrm.org), specifically targets "high-risk" artificial intelligence systems, which include machine-based algorithms that significantly influence decisions regarding employment or employment opportunities (shrm.org). For HR leaders and executives, this shift moves AI oversight from a technical preference to a legal necessity.
What qualifies as a high-risk tool in your stack?
The definition of high-risk is centered on the impact of the decision rather than the complexity of the code. If an algorithm assists in screening resumes, ranking candidates, or predicting employee performance, it likely falls under the scope of SB 24-205 (shrm.org). The mechanism here is the "significant influence" test.
One reading is that this creates a heavy administrative burden for HR departments that use multiple niche vendors. Another reading is that it provides a clear framework for standardizing how we vet technology. We must distinguish between "assistive" tools and "decisive" tools. If a tool merely provides data for a human to review, the risk may be lower, but if the tool automates the rejection of candidates, the risk is high (shrm.org).
What the data does not say is which specific software categories will be targeted first by regulators. However, the law focuses on the outcome of the decision-making process. If the outcome affects a person's ability to secure a job, the tool is high-risk (shrm.org).
The scale of potential liability is tied to the number of individuals affected. A single system impacting 1,000 applicants creates a massive financial risk profile (ai2.work).
Why It Matters
The cost of non-compliance: Assessing the exposure
The financial implications of non-compliance are substantial. If an organization deploys a system that fails to meet disclosure requirements, the exposure is not just a flat fine. Because the law looks at the scale of the impact, a single tool used across a large applicant pool can create massive aggregate liability (ai2.work).
| Scenario Metric | Potential Impact/Requirement |
|---|---|
| Applicant Pool Size | 1,000 individuals (ai2.work) |
| Maximum Potential Exposure | Up to $20 million (ai2.work) |
| Small Business Exemption | Fewer than 50 full-time employees (shrm.org) |
| Assessment Frequency | At least annually (fisherphillips.com) |
When we analyze the mechanism of this liability, we see that it is designed to make the cost of "guessing" on AI safety higher than the cost of auditing it. For large-scale employers, the risk is not just a legal one, but a budgetary one. If a vendor provides a tool that claims to be "bias-free" but fails to provide the documentation required by Colorado law, the employer (the deployer) remains the party facing the primary regulatory risk (ai2.work).
The audit mechanism: Annual vs. Modification-based
The law establishes two distinct triggers for impact assessments. The first is a temporal trigger: assessments must be completed at least once a year (fisherphillips.com). The second is a functional trigger: an assessment must be completed within 90 days after any major modification to the system (fisherphillips.com).
Analysis of this requirement suggests a significant shift in how HR teams must manage vendor relationships. In the past, once a software was implemented, it was often left to run in the background. Under SB 24-205, the "major modification" clause means that if a vendor updates their algorithm to improve accuracy, that update might trigger a new legal requirement for the employer to conduct an assessment (fisherphillips.com).
What the law does not clarify is the definition of a "major modification." Does a minor patch to the user interface trigger an assessment, or only a change to the underlying weights of the algorithm? This ambiguity is where HR teams will likely face the most friction with IT and software vendors. We cannot assume that a vendor's internal update is sufficient for our compliance needs.
Defensive strategies: NIST and ISO frameworks
The law provides a path to an affirmative defense. If an employer can demonstrate that they have aligned their AI use with established frameworks, such as the NIST AI Risk Management Framework or ISO 42001, they may have a legal defense against certain claims (ai2.work).
Analysis of these frameworks shows they are designed to manage risk through documentation and continuous monitoring. Aligning with them is not a "set and forget" task. It requires a bridge between HR (the users of the tool) and IT (the managers of the data). If we use these frameworks, we are essentially creating a paper trail that proves we acted with due diligence (ai2.work).
Alignment with NIST or ISO 42001 serves as an affirmative defense for employers (ai2.work).
Watch For
One reading is that these frameworks are expensive to implement. The counter-reading is that they are significantly less expensive than the $20 million exposure risk identified in the law (ai2.work). For executives, the decision is a trade-off between the cost of rigorous governance and the cost of potential litigation.
The vendor dilemma: Who owns the risk?
A common misconception is that the software vendor is responsible for compliance. While the vendor creates the tool, the employer is the "deployer" of the tool. If the tool is used to make employment decisions, the employer is the one interacting with the Colorado job applicant (shrm.org).
This creates a tension in procurement. HR teams often rely on vendor-run data or self-reported "fairness" metrics. However, a vendor's self-reported assessment may not meet the specific requirements of the Colorado AI Act. We must ask our vendors: "Will your documentation support our annual impact assessment requirement?" (fisherphillips.com).
Analysis of the relationship between deployers and vendors shows a gap in accountability. A vendor may update their model (a "major modification") without notifying the HR team immediately. If that update happens, the 90-day clock for the employer to conduct an assessment starts immediately (fisherphillips.com). Without a contractually mandated notification period, the employer is left vulnerable to a compliance gap.
What HR can do on Monday morning
Compliance is a logistical challenge, not just a legal one. To move from awareness to action, HR teams should execute the following steps:
- Inventory all "high-risk" tools: List every piece of software used in recruitment, screening, or performance management that uses machine-based algorithms to influence decisions (shrm.org).
- Identify small business status: Confirm if your organization has fewer than 50 full-time employees, as this may exempt you from certain requirements (shrm.org).
- Request disclosure documentation: Contact your top three AI vendors and ask for their current impact assessment and their process for notifying you of "major modifications" (fisherphillips.com).
- Schedule the first annual audit: Set a hard date on the calendar for the first mandatory impact assessment for all high-risk systems (fisherphillips.com).
As we move through the first half of 2026, the focus must shift from understanding the law to documenting the process. The Colorado legislature has set the stage, but the burden of proof lies with the employer. We must ensure that our use of technology is as transparent and accountable as the people we are trying to hire.