Texas TRAIGA

The Texas Responsible AI Governance Act (TRAIGA, HB 2060) regulates high-risk AI systems used in consequential decisions affecting Texas residents.

Overview

Aspect	Details
Effective Date	September 1, 2025
Scope	High-risk AI systems in consequential decisions
Penalties	Up to $25,000 per violation
Enforcement	Texas Attorney General

Key Requirements

§ 2056.002: Meaningful Human Review

"A deployer shall ensure that a human makes the final determination... [and] the human review is meaningful"

What "Meaningful" Means:

• Not rubber-stamping AI decisions

• Genuine evaluation of AI recommendations

• Documented override capability

ATHENA Solution:

• Trust Calibration Engine detects rubber-stamping patterns

• Override quality tracking

• Decision time analysis

const result = await athena.calibrate.analyze({
  userId: 'reviewer_123',
  aiRecommendation: 'Deny claim',
  userDecision: 'Deny claim',
  timeToDecisionMs: 500  // Very fast = potential rubber-stamping
});

if (result.pattern === 'rubber_stamping') {
  console.log('Warning: Meaningful review may be lacking');
}

§ 2056.103: Semi-Annual Bias Audits

"A deployer shall conduct a bias audit of each high-risk AI system at least every six months"

Requirements:

• Test for bias across protected classes

• Document findings

• Remediate issues discovered

ATHENA Solution:

• Continuous bias monitoring

• Semi-annual report generation

• Subgroup performance tracking

// Generate semi-annual audit report
const audit = await athena.export.generate({
  template: 'texas_traiga',
  format: 'pdf',
  startDate: '2025-07-01',
  endDate: '2025-12-31'
});

§ 2056.104: Consumer Notification

"Before making a consequential decision... a deployer shall notify the consumer that an AI system was used"

ATHENA Solution:

• Decision logging with timestamps

• Audit trail for compliance proof

TRAIGA Report Contents

1. Bias Audit Summary

• Protected classes analyzed

• Disparities detected

• Statistical significance

• Remediation actions

2. Meaningful Human Review Analysis

• Override rates by reviewer

• Decision time distribution

• Rubber-stamping detection

• Training recommendations

3. Consumer Notifications

• Decisions made in period

• Notification compliance rate

• Any notification failures

4. Semi-Annual Comparison

• Period-over-period trends

• Improvement areas

• Persistent issues

Protected Classes (Texas)

Class	ATHENA Subgroup
Race	ethnicity
Color	ethnicity
National origin	national_origin
Religion	religion
Sex	gender
Age	age_group
Disability	disability_status
Genetic information	genetic_info

Implementation

Step 1: Enable Subgroup Tracking

const result = await athena.bias.detect({
  decisionId: 'dec_123',
  subgroups: {
    gender: 'female',
    ageGroup: '45-54',
    ethnicity: 'hispanic'
  }
});

Step 2: Set Up Alerts

await athena.webhooks.create({
  url: 'https://your-app.com/traiga-alerts',
  events: ['bias.detected'],
  filters: {
    severity: ['high', 'medium']
  }
});

Step 3: Schedule Reports

// Set up automated semi-annual reports
const schedule = cron.schedule('0 0 1 1,7 *', async () => {
  const report = await athena.export.generate({
    template: 'texas_traiga',
    format: 'pdf'
  });
  await emailComplianceTeam(report);
});

Audit Schedule

Period	Deadline
H1 2025	July 31, 2025
H2 2025	January 31, 2026
H1 2026	July 31, 2026
H2 2026	January 31, 2027

Sample Audit Findings

TEXAS TRAIGA SEMI-ANNUAL BIAS AUDIT
Organization: Texas Insurance Co.
Period: July 1 - December 31, 2025

BIAS AUDIT SUMMARY
==================
Protected Classes Analyzed: 8
Subgroups with Disparity: 2
  - age_group:65+ (accuracy disparity: 18%)
  - disability_status:yes (representation disparity: 23%)

Remediation Status: IN PROGRESS
  - Age group: Retraining scheduled for Q1 2026
  - Disability: Additional data collection underway

MEANINGFUL HUMAN REVIEW
=======================
Reviewers Analyzed: 45
Rubber-Stamping Detected: 3 reviewers
  - reviewer_012: 98% follow rate, avg 0.8s decision time
  - reviewer_045: 96% follow rate, avg 1.1s decision time
  - reviewer_078: 97% follow rate, avg 0.9s decision time

Remediation: Training sessions scheduled

COMPLIANCE STATUS: PARTIAL
- Bias audits: COMPLIANT
- Meaningful review: ATTENTION REQUIRED (3 reviewers)
- Consumer notification: COMPLIANT

---

Next: Colorado AI Act