Colorado ADMTA vs EU AI Act vs California: AI Regulations Compared 2026

Three major AI regulatory frameworks are reshaping how developers build and deploy AI in 2026. The Colorado ADMTA replaces the original AI Act with a lighter touch, the EU AI Act’s full enforcement arrives August 2, and California moves forward with multiple targeted bills. Here’s what each requires and how they compare.

Last verified: June 28, 2026

The regulatory landscape at a glance

Jurisdiction	Law	Effective Date	Scope	Penalties
Colorado	ADMTA (SB 26-189)	January 1, 2027	High-risk ADMT that materially influences consequential decisions	Up to $20,000/violation
EU	AI Act	August 2, 2026 (full)	All AI systems placed on EU market or affecting EU persons	Up to 7% global turnover
California	Multiple bills	2026-2027 (phased)	AI safety, workplace, and transparency	$5,000-$25,000/violation

Colorado ADMTA: The reset

The original Colorado AI Act (SB 24-205) was seen as the most comprehensive US state AI law — until the backlash hit. On May 14, 2026, Governor Polis signed SB 26-189, replacing it entirely.

Key changes from the original Colorado AI Act:

• Eliminated: Broad risk assessment programs, impact assessments, general duty to prevent algorithmic discrimination
• Replaced with: Disclosure-based approach, documentation for deployers, public summaries
• Retained: Reasonable care standard, anti-discrimination protections, recordkeeping (3 years)
• New: Consolidated enforcement by the Attorney General, liability limited to intended-use scenarios

What developers must do by January 1, 2027:

1. Inventory AI systems that materially influence consequential decisions
2. Document algorithmic discrimination risks and mitigation methods
3. Publish a public summary of high-risk systems
4. Establish a 90-day reporting process for discovered discrimination
5. Implement a three-year record retention policy

EU AI Act: The gold standard with teeth

The EU AI Act becomes fully applicable on August 2, 2026. Unlike Colorado’s scaled-back approach, the EU maintains comprehensive requirements.

Applicability (August 2, 2026):

• Article 6: Classification rules for high-risk AI systems
• Articles 11-14: Risk management, data governance, technical documentation, transparency
• Article 17: Human oversight requirements
• Articles 19-22: Conformity assessment obligations

What developers must do by August 2, 2026:

1. Classify AI systems under Annex III (high-risk categories)
2. Implement risk management processes (Article 9)
3. Establish data governance for training data (Article 10)
4. Create technical documentation (Article 11)
5. Enable human oversight (Article 14)
6. Register high-risk systems in EU database

Notable developments as of June 2026:

• The Omnibus simplification proposal (April 2026) failed in trilogue — no last-minute relief
• EU AI Office released its final implementation guidelines on June 15
• Each member state must establish at least one AI regulatory sandbox by August 2
• GPAI model providers (OpenAI, Anthropic, Google, Meta) must comply with Code of Practice

California: The fragmented approach

California has pursued multiple targeted bills rather than one comprehensive act.

Active California AI bills (June 2026):

Bill	Focus	Status
Workplace AI Bill	AI in hiring, promotion, and performance evaluation	Passed Assembly, pending Senate
AI Safety Testing Bill	Pre-deployment testing for frontier models	Committee stage
AI Transparency Bill	Disclosure requirements for AI-generated content	Signed into law, effective 2027
Anti-Discrimination Bill	Algorithmic fairness in consumer-facing AI	Proposed

California’s approach is less coordinated but potentially more stringent where it applies. Developers operating in California should monitor all four bills.

Comparison by developer workflow impact

Requirement	Colorado ADMTA	EU AI Act	California
Risk assessment	Light (document discrimination risks)	Heavy (full risk management system)	Medium (safety testing)
Documentation for deployers	Required (summary of risks)	Required (full technical docs)	Not yet required
Human oversight	Not specified	Required for high-risk	Under discussion
Registration	None	EU database registration	None
Cross-border application	Colorado residents only	Applies to any EU-market system	California residents
AI safety testing	Not required	Proportional to risk level	Required for frontier
Enforcement	AG only (deceptive trade practices)	National authorities + AI Office	AG + private right of action

What this means for developers in practice

If you build AI coding tools or agents

Priority: EU AI Act compliance by August 2, 2026

• Most AI coding tools are general-purpose, not high-risk — you may fall under GPAI transparency rules
• If your tool evaluates developer performance, it could be classified as high-risk under Annex III (worker management)

If you build consumer-facing AI apps

Priority: EU AI Act + Colorado ADMTA

• EU: Classification under Annex III (access to essential services, education, etc.)
• Colorado: Disclosure requirements if your app makes or influences consequential decisions
• California: AI transparency disclosure requirements by 2027

If you build foundation models

Priority: EU AI Act (GPAI provisions)

• The EU’s GPAI (General Purpose AI) rules apply to any model with 10^25 or more FLOPs
• Systemic risk designation triggers additional obligations
• Code of Practice compliance is expected by August 2026

The bottom line

The Colorado ADMTA is a significant relaxation from the original AI Act, reflecting industry pushback and legislative realism. The EU AI Act remains the most comprehensive and heavily enforced framework. California is still sorting itself out.

Development teams should:

1. Prioritize EU AI Act compliance (highest penalties, broadest scope)
2. Monitor Colorado ADMTA (precedent for other US states)
3. Watch California (targeted bills could become template for other states)
4. Build documentation processes once (good risk documentation satisfies most jurisdictions)

The divergence between US state and EU approaches means developers face a patchwork. The safest investment: build robust documentation, testing, and transparency practices regardless of jurisdiction.

---

Last verified: June 28, 2026. Sources: Seyfarth analysis of Colorado ADMTA, EU AI Act official timeline, California legislative trackers, Morrison Foerster comparison study.