Free Reference · 2026-2028

AI Compliance Deadlines & Documentation Reference - 2026-2028

Five frameworks. Key 2026-2028 AI compliance milestones. One reference sheet.

This page is an informational reference, not legal advice or a compliance determination. Dates, enforcement timelines, and rule status can change. Review the linked primary sources and consult qualified legal counsel before relying on this page for compliance decisions.

Official starting points: EU AI Act text, EU AI Act implementation timeline, European Commission Article 50 Code of Practice, Colorado SB 26-189, and Colorado AG AI rulemaking.

Try the Free Classifier PDF download paused

PDF download is paused while the reference is updated for recent Colorado and EU AI Act timeline changes. Use this page as the current public reference.

August 2, 2026
EU AI Act Article 50 transparency obligations
December 2, 2026
Certain Article 50(2) pre-existing-system transitions
January 1, 2027
Colorado ADMT Act / SB 26-189
January 1, 2027
CPRA ADMT
December 2, 2027
EU AI Act Annex III high-risk systems

NYC LL144 and Illinois AIVIA/HB 3773 - already in effect. EU Annex I product-embedded high-risk system obligations: August 2, 2028, where applicable.

Colorado Automated Decision-Making Technology Act / SB 26-189
January 1, 2027
Status: SB 26-189 repealed and re-enacted Colorado's prior AI provisions. Rules are pending; confirm final requirements before relying on this summary.

Developers and deployers of automated decision-making technology used to materially influence consequential decisions affecting Colorado consumers. Final scope and implementation details may depend on pending rules.

  • Technical and system documentation for covered automated decision-making technology
  • Risk management and governance records
  • Consumer notice and post-adverse outcome disclosure records
  • Meaningful human review or reconsideration records, where applicable
  • Record retention and review history
  • Rulemaking and implementation tracking before January 1, 2027
Colorado Consumer Protection Act enforcement may apply. Penalty amounts, cure opportunities, and enforcement outcomes depend on final rules, the facts, enforcement authority, and applicable legal analysis.
Pending Rules
Treat this as a tracking milestone, not a settled final-rule checklist.
EU AI Act - Article 50 and High-Risk Milestones
2026-2028
Status: Active regulation with phased application. Article 50 transparency obligations apply August 2, 2026. Certain Article 50(2) pre-existing-system transitions run to December 2, 2026. Annex III high-risk system obligations apply December 2, 2027. Annex I product-embedded high-risk system obligations apply August 2, 2028, where applicable.

Providers and deployers of AI systems serving EU users, including systems subject to Article 50 transparency obligations and systems in Annex III high-risk domains (biometrics, critical infrastructure, education, employment, essential services/credit scoring, law enforcement, migration, justice/democracy). Timing depends on the obligation and system category.

  • Annex IV technical documentation (9 sections: system description, design specs, data requirements, human oversight, predetermined changes, validation/testing, risk management, standards mapping, post-market monitoring)
  • Risk management system (Article 9)
  • Data governance procedures (Article 10)
  • Conformity assessment (self-assessment or third-party per Annex VI/VII)
  • EU database registration (Annex VIII fields)
  • Post-market monitoring plan (Article 72)
  • Serious incident reporting protocol (Article 73)
  • CE marking and EU declaration of conformity
  • Transparency disclosures and implementation records, where Article 50 applies
Up to €35M or 7% of global turnover for prohibited practices. Up to €15M or 3% for high-risk non-compliance. Penalty amounts and enforcement outcomes depend on the facts, enforcement authority, and applicable legal analysis.
CPRA ADMT (California)
January 1, 2027
Status: Regulations effective January 1, 2026. Risk assessment compliance begins January 1, 2026. Submission deadlines are phased: April 1, 2028 / 2029 / 2030 depending on annual revenue.

Any CCPA "business" (for-profit, doing business in CA, meeting revenue/data volume thresholds) using automated decision-making technology for "significant decisions" (financial/lending, housing, education, employment, healthcare) affecting California consumers.

  • Pre-use consumer notice (purpose, how ADMT works, outputs, alternative process)
  • Opt-out mechanism (or documented exception: human appeal, admission/hiring, work allocation)
  • Consumer access rights response procedures
  • Risk assessment report (purpose, impacts, safeguards, governance signoff)
  • Risk assessment 3-year review cycle with 45-day material change updates
  • Annual metrics compilation and disclosure (if processing PI of 10M+ consumers)
  • Anti-dark pattern UI testing documentation
  • Executive management attestation for risk assessment submissions
Standard CCPA enforcement - up to $7,500 per intentional violation. Penalty amounts and enforcement outcomes depend on the facts, enforcement authority, and applicable legal analysis.
Key Nuance
ADMT = computation + personal information + replaces or substantially replaces human decision-making. Advertising is explicitly excluded. Human involvement (3-part AND test) can take you out of scope.
NYC Local Law 144
Already Enforced
Status: Active DCWP enforcement. Effective July 5, 2023.

Employers and employment agencies using automated employment decision tools (AEDTs) for hiring or promotion decisions in New York City.

  • Annual independent bias audit (selection rate and impact ratio by race/ethnicity and sex)
  • Published bias audit summary on employer's website
  • Candidate notice (at least 10 business days before AEDT use)
  • Data type disclosure (what data the AEDT collects and analyzes)
  • Alternative process disclosure (how candidates can request alternative selection)
  • Records retention for bias audits (minimum 4 years under DCWP rules)
$375–$1,500 per violation (first offense: $500). Each day of non-compliance with notice = separate violation. Each person not notified = separate violation. Penalty amounts and enforcement outcomes depend on the facts, enforcement authority, and applicable legal analysis.
Illinois AI Employment Laws (AIVIA + HB 3773)
Already Enforced
Status: Both laws active. AIVIA: January 1, 2020. HB 3773: January 1, 2026.

AIVIA - any employer using AI to analyze video interviews in Illinois. HB 3773 - any employer using AI for employment decisions (screening, evaluation, discipline, termination) in Illinois.

  • Pre-interview notice (AI analysis disclosed before interview)
  • Applicant consent (affirmative consent required before AI analysis)
  • Video deletion process (within 30 days of applicant request)
  • AI provider attestation or supporting records
  • Notice to employees and applicants when AI is used in employment decisions
  • Prohibition on using AI in a way that results in unlawful discrimination based on protected characteristics
  • Demographic data collection and reporting to IDCEO - applies only when an employer relies solely on AI video interview analysis to determine whether an applicant receives an in-person interview
  • Bias monitoring measures
Enforced through Illinois Human Rights Act and Illinois Department of Labor. Civil penalties vary. Penalty amounts and enforcement outcomes depend on the facts, enforcement authority, and applicable legal analysis.
Cross-Framework Insight

Many of these frameworks require overlapping documentation. A single AI system used in hiring may raise review questions across multiple frameworks. The core artifacts - system description, risk assessment, notice mechanisms, bias controls - overlap significantly. The challenge is mapping one set of documentation to five different legal frameworks with different terminology and citation requirements.

Screen selected frameworks that may be relevant for review in 2 minutes.

Free multi-framework screener - no signup needed.

Take the Free Screener →

Need to produce the actual documentation?

Berly helps organize AI system records, supporting evidence, review notes, and exportable documentation packages across selected AI governance frameworks.

Learn More →

AI Documentation Crosswalk →

See something outdated or unclear? Send a correction or practitioner note.