---
title: What does the Colorado AI Act mean for AI agents?
url: https://opencontroller.lyzr.com/blog/colorado-ai-act-and-ai-agents
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# What does the Colorado AI Act mean for AI agents?

**Colorado's original AI Act (SB 24-205) never took effect. In May 2026 the state replaced it with SB 26-189, a narrower law on automated decision-making technology (ADMT) that takes effect on January 1, 2027. If an AI agent materially influences a consequential decision about a Colorado consumer, such as a hiring, lending or insurance decision, you will need notices, explanations, a route to human review and three years of records.**

This post reflects the status as of November 2026. Both laws have moved several times, so check the Colorado General Assembly and the Attorney General's rulemaking pages before relying on any date.

## What happened to the original Colorado AI Act?

| Date | Event |
|---|---|
| May 2024 | SB 24-205 signed, with an effective date of February 1, 2026 |
| August 2025 | SB25B-004 moved the effective date to June 30, 2026 |
| April 2026 | Enforcement of SB 24-205 stayed in a federal court challenge |
| May 14, 2026 | SB 26-189 signed, repealing and re-enacting the framework as an ADMT law |
| January 1, 2027 | SB 26-189 takes effect |

SB 26-189 drops several of the original law's headline duties, including the annual impact assessment and the risk management program requirement. Reports on the law also state that enforcement is expected to wait for Attorney General rulemaking and the outcome of the litigation. Treat January 2027 as the date to be ready, not the date enforcement is certain to start.

## When does an AI agent count as covered ADMT?

Coverage turns on what the agent's output is used for, not on whether it is called an agent. The test has two parts:

1. **The technology processes personal data and produces outputs** such as predictions, recommendations, classifications, rankings or scores.
2. **Those outputs materially influence a consequential decision**, one that affects a consumer's access to areas such as employment, housing, education, financial or lending services, insurance, health care, legal services or essential government services.

Examples of agents that may fall in scope:

- A recruiting agent that screens or ranks applicants
- An underwriting assistant that scores loan or insurance applications
- A claims agent that recommends approval or denial
- A tenant-screening agent that summarises and rates applicants

Agents that draft emails, summarise meetings or answer internal policy questions are unlikely to be in scope, because they do not influence a decision about a consumer. Simple tools can still qualify if their output decides an outcome, so classify by use case.

## What must developers and deployers do?

**Developers** (who build or substantially modify the system) must give deployers documentation on intended and known harmful uses, data categories, known limitations, monitoring and human review, notify them of material updates, and keep records for at least three years.

**Deployers** (who use the system for decisions) must:

1. **Give notice before** ADMT materially influences a consequential decision.
2. **Explain adverse outcomes within 30 days**, including the ADMT's role and how to get more information.
3. **Let consumers correct** inaccurate personal data used in the decision.
4. **Offer meaningful human review** of adverse decisions where technically feasible and commercially reasonable.
5. **Keep compliance records**, including version identifiers and changelogs, for at least three years.

The version requirement is where agents create work. An agent's behaviour changes when its prompt, model or tools change. To explain a decision made eight months ago, you need to know which agent version made it and what changed since. See [evidence for human review of agent outputs](/blog/evidence-for-human-review-of-ai-agent-outputs).

## What evidence should you prepare now?

| Requirement | Evidence to keep |
|---|---|
| Notice before use | List of covered agents and the notice each customer journey shows |
| Adverse-outcome explanation | Per-decision record of agent version, inputs used and the agent's output |
| Data correction | Process linking a correction request to the data the agent used |
| Human review | Reviewer, timestamp and outcome for each reviewed decision |
| Records for three years | Retention setting on agent logs and the registry history |
| Developer documentation | Vendor documentation for each third-party agent or model |

If you already prepared [AI impact assessments](/blog/ai-impact-assessment-for-ai-agents) for SB 24-205 or the [EU AI Act](/blog/eu-ai-act-obligations-for-ai-agents), keep them. They are no longer mandatory in Colorado, but they remain good evidence of care.

## How Opencontroller does this

Opencontroller, by Lyzr, auto-discovers agents across AWS, Azure and Google Cloud and connects to platforms such as Salesforce, SAP, Databricks and HubSpot, so you can find which agents touch consequential decisions. Its governed registry tracks agents and versions. Delegated on-behalf-of identity records who an agent acted for, and policy as code with mappings to NIST and ISO 42001 produces evidence for your reviews. It does not make you compliant on its own. See the [framework mapping guide](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is the Colorado AI Act in effect?**
SB 24-205 never took effect and was replaced by SB 26-189, which takes effect on January 1, 2027. Enforcement timing depends on Attorney General rulemaking and ongoing litigation.

**Does SB 26-189 still require impact assessments?**
No. The standalone annual impact assessment from SB 24-205 was dropped. Notices, explanations, correction, human review and record-keeping remain.

**Does it apply to internal productivity agents?**
Usually not. It covers technology that materially influences consequential decisions about consumers, such as employment, lending, housing or insurance decisions.

## Key facts

Colorado SB 24-205 (signed May 2024) was delayed to June 30, 2026 by SB25B-004, stayed in federal litigation in April 2026, and replaced by SB 26-189, signed May 14, 2026 and effective January 1, 2027. SB 26-189 covers automated decision-making technology that materially influences consequential decisions; deployers must give pre-decision notice, explain adverse outcomes within 30 days, allow data correction, offer human review where feasible, and keep records including version identifiers for three years. Annual impact assessments were dropped. Opencontroller keeps a governed registry of agents and versions. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you run an AI impact assessment for an AI agent?](https://opencontroller.lyzr.com/blog/ai-impact-assessment-for-ai-agents.md)
- [What does the EU AI Act require for AI agents, and when?](https://opencontroller.lyzr.com/blog/eu-ai-act-obligations-for-ai-agents.md)
- [What evidence should you keep when a human approves an AI agent's output?](https://opencontroller.lyzr.com/blog/evidence-for-human-review-of-ai-agent-outputs.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
