Who this is for: HR and employment counsel, healthcare compliance teams, AI product managers, and GRC architects responsible for Connecticut AI obligations across employment, healthcare, and consumer-facing AI systems.

Signed May 27, 2026. The Connecticut AI Responsibility and Transparency Act (CART Act, SB 5) is the most comprehensive US state AI law. AEDT discrimination liability begins October 1, 2026. Employer notice requirements begin October 1, 2027. This is a separate law from Connecticut SB 2 (AI Accountability Act). See the SB 2 crosswalk for the earlier, narrower law.

Contents

1. Overview 2. Compliance Timeline 3. Key Obligations 4. Obligation-to-Procedure Mapping 5. Detailed Procedure Cards 6. Quick Reference 7. Quick Start 8. References

1. Overview

The CART Act covers four distinct AI domains in a single law, making it the broadest US state AI regulation enacted to date:

The Act applies to employers doing business in Connecticut, employers with Connecticut employees, and employers accepting job applications from Connecticut residents. The employment provisions have the broadest reach and earliest enforcement dates.

2. Compliance Timeline

DateWhat Takes EffectWho It Affects
May 27, 2026Law signed by Governor LamontAll covered entities
Oct 1, 2026AEDT discrimination liability begins; employer use of AEDT is not a defense to discrimination complaintsAll employers using AI in employment decisions
Oct 1, 2026WARN Act AI disclosure: employers must state whether layoffs are "related to the employer's use of artificial intelligence"Employers filing plant closing or mass RIF notices
Oct 1, 2027Full employer notice requirements: disclose AI use in hiring, provide explanation of AI's role, offer human review optionEmployers with CT employees or CT applicants

3. Key Obligations

ObligationDomainRequirement
AEDT Discrimination LiabilityEmploymentAI use is not a defense to discrimination complaints. Employers bear responsibility for AI-driven discriminatory outcomes.
Employer AI DisclosureEmploymentNotify candidates and employees when AEDT is used in employment decisions. Provide explanation of AI's role.
Human Review OptionEmploymentOffer affected individuals the option of human review for AI-driven employment decisions.
WARN Act AI DisclosureEmploymentDisclose whether mass layoffs are related to AI or technological change in WARN Act filings.
Healthcare AI TransparencyHealthcareDisclose AI involvement in clinical decisions, diagnosis support, and treatment recommendations.
Companion Chatbot SafetyOnline SafetySafety measures for AI companion chatbots interacting with minors.
Consumer NotificationConsumerNotify consumers when consequential decisions are made using automated systems.

4. Obligation-to-Procedure Mapping

CART Act ObligationSWT3 ProcedureWhat It WitnessesEvidence Produced
AEDT DiscriminationAI-FAIR.1Bias detection and fairness attestationFactor A: protected attribute, Factor B: metric result, Factor C: threshold
AEDT DiscriminationAI-DPIA.1Impact assessment completionFactor A: assessment scope, Factor B: risk rating, Factor C: review authority
Employer AI DisclosureAI-TRANS.1Transparency report generationFactor A: report type, Factor B: coverage, Factor C: publication status
Employer AI DisclosureAI-EXPL.1Explanation generation and deliveryFactor A: explanation method, Factor B: confidence, Factor C: factors cited
Human Review OptionAI-HITL.1Human-in-the-loop decision verificationFactor A: decision type, Factor B: reviewer hash, Factor C: override authority
Healthcare TransparencyAI-INF.1Inference provenanceFactor A: model identifier, Factor B: provider, Factor C: clearing level
Healthcare TransparencyAI-HITL.1Human oversight of clinical AIFactor A: decision type, Factor B: reviewer hash, Factor C: override authority
Companion SafetyAI-SAFE.1Safety testing and validationFactor A: test type, Factor B: pass rate, Factor C: coverage
Companion SafetyAI-GRD.1Guardrail configuration attestationFactor A: guardrail type, Factor B: rule count, Factor C: enforcement mode
Consumer NotificationAI-TRANS.1Transparency disclosureFactor A: report type, Factor B: coverage, Factor C: publication status

5. Detailed Procedure Cards

AI-FAIR.1

Bias Detection and Fairness Attestation

CART Act requires: Employers are liable for discriminatory outcomes from AEDT. Using AI is "not a defense to a complaint alleging a discriminatory practice." This effectively requires employers to test for and mitigate algorithmic bias before and during deployment of AI hiring tools.

How SWT3 addresses it: The witness_fairness() call records which protected attribute was tested (gender, age, race, disability), the fairness metric result, and the threshold applied. Regular AI-FAIR.1 anchors demonstrate continuous bias monitoring, not just pre-deployment testing.

What to show the examiner

AI-FAIR.1 anchors at regular intervals prove bias monitoring is active. Factor A identifies the protected attribute. Factor B records the metric result (demographic parity, equalized odds, disparate impact ratio). If any Factor B value exceeds the threshold in Factor C, cross-reference with remediation records to show the employer took corrective action. The absence of AI-FAIR.1 anchors during AEDT operation is itself evidence of non-compliance.

AI-EXPL.1

Explanation Generation

CART Act requires: Beginning October 1, 2027, employers must provide an "explanation of the AI's role" in employment decisions to affected candidates and employees. The explanation must be sufficient for the individual to understand how AI influenced the outcome.

How SWT3 addresses it: The witness_explanation() call records the explanation method used, the confidence score, and the factors cited in the explanation. Each explanation delivery produces an anchor proving the employer fulfilled the disclosure obligation for that specific decision.

What to show the examiner

AI-EXPL.1 anchors prove explanations were generated and delivered. Factor A identifies the method. Factor C lists the factors cited, which should correspond to legitimate job-related criteria. The ratio of AI-INF.1 (inference) to AI-EXPL.1 (explanation) anchors shows what percentage of AI-driven decisions received explanations. For CART Act compliance, this ratio should approach 1:1 for consequential employment decisions.

AI-HITL.1

Human-in-the-Loop Decision Verification

CART Act requires: Employers must offer affected individuals the option of human review for AI-driven employment decisions. For healthcare AI, human oversight is required for clinical decision support and diagnostic recommendations.

How SWT3 addresses it: The witness_hitl() call records the decision type, a hash of the reviewer's identity, and the override authority level. This creates evidence that human review was available and, when exercised, that a qualified individual participated.

What to show the examiner

AI-HITL.1 anchors prove human review infrastructure exists. For employment: show that human review was offered (AI-TRANS.1 disclosure) and exercised when requested (AI-HITL.1 anchors with matching decision types). For healthcare: show that every clinical AI output has a corresponding AI-HITL.1 anchor from a qualified clinician.

AI-TRANS.1

Transparency Report Generation

CART Act requires: Multiple transparency obligations: employer disclosure of AI use in hiring, WARN Act AI disclosure, consumer notification for consequential decisions, and healthcare AI disclosure. All require proactive notification to affected individuals.

How SWT3 addresses it: The witness_transparency() call records the report type, coverage, and publication status. Each disclosure event produces an anchor proving the organization fulfilled its notification obligation at a specific time for a specific audience.

What to show the examiner

AI-TRANS.1 anchors prove transparency disclosures were made. Factor A identifies the disclosure type (employment notice, WARN filing, consumer notification, healthcare disclosure). Factor C records publication status. For employment: AI-TRANS.1 anchors should appear before or concurrent with AI-INF.1 anchors for the same decision process, proving disclosure preceded the AI-driven outcome.

AI-DPIA.1

Impact Assessment Completion

CART Act requires: The AEDT discrimination liability provisions effectively require employers to conduct impact assessments before deploying AI in employment decisions. Without documented assessment, employers cannot demonstrate they took reasonable steps to prevent discriminatory outcomes.

How SWT3 addresses it: The witness_dpia() call captures the assessment scope, risk rating, and review authority. The anchor timestamp proves the assessment was completed before the AEDT was deployed in production.

What to show the examiner

AI-DPIA.1 anchor timestamps must predate the first AI-INF.1 anchor for the same AEDT system, proving assessment before deployment. Factor B records the risk rating. Factor C identifies who approved the assessment. Cross-reference with AI-FAIR.1 anchors to show that bias testing was part of the impact assessment process.

6. Quick Reference

Examiner QuestionWhere to Look
Do you test your hiring AI for bias?AI-FAIR.1 anchors at regular intervals. Factor A identifies protected attributes tested. Factor B shows metric results. Continuous anchors prove ongoing monitoring, not just pre-deployment testing.
Did you conduct an impact assessment before deploying AEDT?AI-DPIA.1 anchor timestamp must predate the first AI-INF.1 anchor for the same system. Factor C identifies the approving authority.
Do you notify candidates about AI use in hiring?AI-TRANS.1 anchors with Factor A = "employment notice." Anchors should appear before or concurrent with AI-INF.1 for the same decision process.
Can candidates request human review?AI-HITL.1 anchors prove the capability exists. The ratio of human review requests to AI-driven decisions shows how often the option is exercised.
Did you disclose AI in your WARN Act filing?AI-TRANS.1 anchors with Factor A = "WARN filing." The anchor provides independently verifiable proof that the AI disclosure was included.
How do you ensure companion chatbot safety?AI-SAFE.1 anchors for safety testing. AI-GRD.1 anchors for guardrail configuration. Factor B in AI-GRD.1 shows rule count; Factor C shows enforcement mode (blocking vs advisory).

7. Quick Start

# Install the SDK
pip install swt3-ai

# Initialize with the NIST AI RMF profile (covers all CART Act obligation areas)
swt3 init --profile nist-ai-rmf --tenant YOUR_TENANT

# Run the demo to see witness anchors generated
python -m swt3_ai.demo

# Or use TypeScript
npm install @tenova/swt3-ai
npx swt3-init --profile nist-ai-rmf

Full SDK documentation: sovereign.tenova.io/docs

Create a free account: sovereign.tenova.io/signup

8. References