Understanding the Colorado Artificial Intelligence Act (SB 24-205)
The passage of Colorado SB 24-205 represents a watershed moment in the landscape of AI governance within the United States. Following closely on the heels of the European Union AI Act, Colorado is the first US state to implement a risk-focused framework that regulates how private businesses and governmental entities develop and deploy artificial intelligence models.
The Definition of "High-Risk AI Systems"
Unlike loose definitions of automation, the Colorado AI Act targets software that makes or serves as a substantial factor in making consequential decisions. These decisions include anything impacting critical consumer aspects such as access to employment, hiring or termination processes, academic evaluations and enrollment opportunities, housing access, financial loans, essential utility services, healthcare interventions, or legal representation.
Key Developer vs. Deployer Obligations
The law establishes separate duties of care to prevent algorithmic discrimination:
- Developers: Those who build or alter models must provide rigorous documentation to deployers, detailing model training criteria, risk mitigations, validation checks, and potential limitations. They are also required to report detected biases to the Attorney General within 90 days.
- Deployers: Those who utilize high-risk systems in active operations must build a dedicated risk management program (relying on industry frameworks like NIST). Additionally, deployers must complete comprehensive annual Algorithmic Discrimination Impact Assessments (ADIAs).
The Critical Small-Business Safe Harbor
Recognizing the impact of stringent regulations on growing startups, SB 24-205 outlines robust carve-outs for deployers with fewer than 50 employees. Eligible small organizations are exempted from maintaining formal NIST risk policies and performing annual third-party ADIAs, provided they retain a detailed inventory of active systems, prevent active discrimination, and honor consumer rights to contest automated decisions.
Preparing for the February 2026 Mandate
With full implementation on February 1, 2026, companies must begin early system cataloging and risk categorization. Ensuring that your engineering, compliance, and legal teams are aligned using structured assessment tools will mitigate operational interruptions and secure your standing under state-level regulatory review.