Texas now has an in-force AI statute that reaches any business offering AI-enabled products or services to Texans, but most of its duties turn on intent and on who the deployer is, which makes it easy to misread in both directions. The Texas Responsible Artificial Intelligence Governance Act (TRAIGA), enacted as House Bill 149 of the 89th Legislature, took effect on 1 January 2026. It does not copy the risk-tiered model of the EU AI Act or the Colorado AI Act: it sets a short list of prohibited practices, disclosure duties for government agencies and health care providers, Attorney General enforcement with a cure period, a regulatory sandbox and an advisory council. This guide sets out what the enrolled text says, what status could be confirmed as at 24 October 2026, and how security, privacy and AI governance teams can turn it into controls and evidence.
Key takeaways
- It is in force and broad in reach. TRAIGA applies to anyone who does business in Texas, offers products or services used by Texas residents, or develops or deploys an AI system in Texas, and it has applied since 1 January 2026.
- Most private-sector prohibitions are intent based. Manipulation, constitutional-rights and discrimination bans require intent, and the Act states that disparate impact alone does not show intent to discriminate.
- Disclosure duties fall on government and health care. State agencies must tell consumers they are dealing with AI, and health care providers must disclose AI use in services or treatment by the first date of service.
- Enforcement is Attorney General only, with a 60-day cure. There is no private right of action; penalties run from $10,000 for curable violations to $200,000 for uncurable ones, plus up to $40,000 per day for continuing violations.
- Documented testing is a legal asset. The Act gives defences where a violation is found through red teaming or adversarial testing, or where the organisation substantially complies with the NIST AI RMF Generative AI Profile or a similar recognised framework.
Scope: who and what is covered
TRAIGA adds new Title 11 provisions to the Texas Business and Commerce Code (Chapters 551 to 554) and amends related biometric, privacy and government code sections. The scope definitions are short, which is why they matter.
| Term | Section | What the enrolled text says | Practical reading |
|---|---|---|---|
| Artificial intelligence system | § 551.001(1) | A machine-based system that infers from its inputs how to generate outputs that can influence physical or virtual environments | Covers generative models, classifiers, recommenders and AI agents; not limited to "high-risk" systems |
| Consumer | § 551.001(2) | A Texas resident acting in an individual or household context, not a commercial or employment context | Employee-facing and B2B uses fall outside consumer-based provisions, but not outside the prohibitions |
| Developer | § 552.001(2) | A person who develops an AI system offered or provided in Texas | Model builders and vendors embedding AI into products |
| Deployer | § 552.001(1) | A person who deploys an AI system for use in Texas | Any organisation putting an AI system or agent into use for Texans |
| Governmental entity | § 552.001(3) | State and political subdivision units, excluding certain hospital districts and institutions of higher education | Target of the social scoring, biometric and disclosure duties |
| Applicability | § 551.002 | Persons doing business in Texas, providing products or services used by Texas residents, or developing or deploying AI in Texas | Out-of-state companies with Texas users are in scope |
Two features narrow the practical reach. First, the heaviest public-sector duties (disclosure, social scoring and biometric identification) apply to governmental entities, not private businesses. Second, the Attorney General's page on the Act notes a carve-out from the discrimination provision for certain insurers and federally insured financial institutions that comply with, or are regulated under, other laws. For the wider landscape of how Texas compares with other jurisdictions, see the Global AI Regulation Map.
Key obligations
Prohibited practices (Chapter 552, Subchapter B)
| Section | Prohibition | Who it binds | Intent test |
|---|---|---|---|
| § 552.052 | Developing or deploying AI intended to incite or encourage self-harm (including suicide), harm to another person, or criminal activity | Any person | Intended purpose |
| § 552.053 | Social scoring that leads to detrimental or disproportionate treatment or infringes legal rights | Governmental entities | Use based |
| § 552.054 | Using AI to uniquely identify a person by biometric data, or gathering images or media without consent, where this infringes legal rights | Governmental entities | Use based; a violation of § 503.001 is also a violation here |
| § 552.055 | Developing or deploying AI with the sole intent of infringing rights under the US Constitution | Any person | Sole intent |
| § 552.056 | Developing or deploying AI with intent to unlawfully discriminate against a protected class | Any person (with financial and insurance carve-outs) | Intent; disparate impact alone is not sufficient (§ 552.056(c)) |
| § 552.057 | Developing or distributing AI with the sole intent of producing prohibited sexual material, sexual deepfakes, or sexual text conversations impersonating a minor | Any person | Sole intent |
Because most private-sector bans hinge on intent, the evidence that matters is the documented purpose of a system: design documents, acceptable use policies, system prompts, guardrail configurations and test records. An organisation that cannot show what a model or agent was designed to do is in a weaker position when the Attorney General asks.
Disclosure duties (§ 552.051)
- Governmental agencies: must disclose to consumers, before or at the time of interaction, that they are interacting with an AI system. The disclosure must be clear and conspicuous, in plain language, and must not use a dark pattern.
- Health care services: a provider using AI in a service or treatment must disclose it to the recipient or their personal representative no later than the date the service is first provided, or as soon as reasonably possible in an emergency (§ 552.051(f)).
- Private businesses generally: TRAIGA does not create a general chatbot disclosure duty for private businesses outside health care. Other state laws may, see the Utah AI Policy Act Guide and the California AI Laws Guide.
Related amendments that touch data and identity
- Biometric identifiers (§ 503.001): the Act adds subsection (b-1) on images found online, an exemption in subsection (e) for certain AI training uses unless the system is used to uniquely identify a specific person, and subsection (f) applying penalties where biometric data captured for AI training is later used for commercial purposes.
- Processors (§ 541.104(a)): processors under the Texas Data Privacy and Security Act must help controllers meet obligations for personal data collected, stored and processed by AI systems. See the US State Privacy Laws Guide.
- State agency inventories: amendments to Government Code §§ 2054.068(b) and 2054.0965(b) require agencies to report AI use and to include AI systems in their information resources reviews, and § 325.011 adds AI use to Sunset review criteria.
Timeline and current status
| Date | Event | Source |
|---|---|---|
| 2025 (89th Legislature, Regular Session) | HB 149 enrolled | Enrolled bill text |
| 1 January 2026 | Act takes effect (Act Sec. 10) | Enrolled text; Texas Attorney General |
| By 1 September 2026 | Attorney General must post an online complaint mechanism (Act Sec. 8) | Enrolled text |
| October 2026 | The Attorney General's TRAIGA page links a live "File An AI Complaint" portal | Texas Attorney General |
Status as at 24 October 2026: the Attorney General describes the Act as having taken effect on 1 January 2026, and the complaint mechanism is live. We could not confirm from an official page whether the Department of Information Resources has admitted participants to the sandbox, or whether the council has published reports, so this guide does not state either. The Texas Legislature was not in regular session in 2026; check for special-session amendments before relying on this page.
Federal context: an 11 December 2025 White House executive order directed the US Attorney General to set up an AI Litigation Task Force to challenge state AI laws considered unlawful and called for a federal framework that would preempt some of them. According to the White House fact sheet, the order does not itself preempt state law. TRAIGA remains in force unless a court or Congress says otherwise.
Enforcement and penalties
The Attorney General has exclusive enforcement authority (§ 552.101). There is no private right of action, so consumers cannot sue under TRAIGA directly, but they can complain through the Attorney General's portal (§ 552.102), and a complaint can start an investigation.
Investigation and cure
- Complaint received through the online mechanism.
- Civil investigative demand (§ 552.103): the Attorney General may request a description of the system's purpose, the types of data used to train it, performance metrics, known limitations, and post-deployment monitoring and safeguards.
- Written notice (§ 552.104): the Attorney General cannot sue before the 60th day after notice. No action follows if, within that window, the person cures the violation and submits a written statement with supporting documentation and any policy changes made to prevent recurrence.
- Civil action (§ 552.105): if the violation is not cured, or the cure statement is breached.
| Category | Penalty range | Section |
|---|---|---|
| Curable violation, or breach of a cure statement | $10,000 to $12,000 per violation | § 552.105(a) |
| Uncurable violation | $80,000 to $200,000 per violation | § 552.105(a) |
| Continuing violation | $2,000 to $40,000 per day | § 552.105(a) |
| Other relief | Injunction, attorney's fees, court costs and investigative expenses | § 552.105(b) |
| Licensed persons | State licensing agency may suspend, place on probation or revoke a licence, and impose up to $100,000, after a § 552.105 finding and on the Attorney General's recommendation | § 552.106 |
Defences and presumptions
- Rebuttable presumption of reasonable care for the defendant (§ 552.105(c)).
- Third-party misuse: no liability where another person used the system in a prohibited way.
- Self-discovery: no liability where the violation was discovered through user feedback, testing including adversarial or red-team testing, or compliance with state agency guidelines.
- Framework compliance: substantial compliance with the NIST "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile" (NIST AI 600-1) or a similar nationally or internationally recognised framework. See the NIST AI RMF Guide.
- Undeployed systems: no civil penalty action for a system that has not been deployed (§ 552.105(f)).
The practical consequence: the CID list (purpose, training data, metrics, limitations, monitoring) and the defence list (testing, framework alignment) together describe the evidence pack every Texas-facing AI deployment should have ready before anyone asks.
The regulatory sandbox and the AI council
| Feature | Sandbox (Chapter 553) | Texas Artificial Intelligence Council (Chapter 554) |
|---|---|---|
| Administered by | Department of Information Resources, in consultation with the council (§ 553.051) | Administratively attached to the Department of Information Resources (§ 554.001(b)) |
| Entry or membership | Approval by the department and any applicable agency (§ 553.052) | Seven members: three appointed by the governor, two by the lieutenant governor, two by the speaker; governor appoints the chair; staggered four-year terms (§ 554.002) |
| Duration | Testing for up to 36 months, extendable for good cause (§ 553.053) | Standing body |
| Reporting | Participants report quarterly (§ 553.102); the department reports annually to the Legislature (§ 553.103) | Advisory; cannot adopt binding rules (§ 554.103) |
| Limits | The Subchapter B prohibitions cannot be waived (§ 553.051(e)) | No enforcement powers |
When to use the sandbox: it suits developers who want to test an AI product with the department's and any relevant agency's approval, and who can commit to quarterly reporting. It does not offer relief from the prohibited practices, so it is not a route to deploy something the Act forbids. Smaller firms testing a single novel product are the natural participants; large firms with mature programmes usually gain more from building the evidence behind the § 552.105 defences.
What it means for AI agents and non-human identities
TRAIGA never mentions agents or machine identities, but its enforcement mechanics land on them. An AI agent that takes actions for Texas consumers is an AI system that a deployer has deployed, and a civil investigative demand can ask about its purpose, limitations and post-deployment monitoring.
- Purpose is evidence: agent scopes, tool permissions and system prompts record what the agent was intended to do, which is the core of the intent tests in §§ 552.052 to 552.057. Over-broad tool access makes "intended purpose" harder to show. See the AI Agent Authorisation Guide.
- Each agent needs an owner and an identity: a cure statement under § 552.104 has to name the policy changes made. That is only possible if every agent, API key and service account that backs it has a named owner. See the Identity Governance for AI Agents Guide.
- Logs carry the defence: the self-discovery defence depends on showing that testing or monitoring found the problem. Agent action logs tied to a distinct workload identity, not a shared human account, make that provable. See the AI Agent Observability and Incident Response Guide.
- Biometric data in agent pipelines: agents that process images or voice for identification can trigger § 503.001 duties, including the new rule on commercial use of data captured for AI training.
- Liability for what the agent says: tribunals have already held organisations to statements made by their bots; the Air Canada chatbot ruling shows why a deployer cannot treat an agent as a separate actor.
Compliance steps
- Determine applicability: list every AI system, model and agent used by or offered to Texas residents, marking consumer-facing, employee-facing and government-contract uses. Use the AI Register and AI Inventory Template.
- Screen against § 552.052 to § 552.057: for each system, record the intended purpose and confirm none of the six prohibited purposes applies. Pay special attention to generative image, voice and companion features (§ 552.057).
- Apply disclosure where required: health care providers and public bodies implement clear, plain-language AI notices before or at interaction; test the user interface for dark patterns.
- Build the CID evidence pack: per system, keep purpose, training data types, performance metrics, known limitations and post-deployment monitoring in one versioned record, so a 60-day cure window is not spent finding documents.
- Adopt a recognised framework: map your AI governance to NIST AI 600-1 and the AI RMF functions (Govern, Map, Measure, Manage) and record the mapping. See the AI Governance Guide.
- Run and record adversarial testing: schedule red-team exercises before release and after material changes, track findings to closure, and keep reports. See the Red Teaming AI Agents for Identity Abuse.
- Prepare a cure playbook: define who receives an Attorney General notice, who can switch off or roll back a model or agent version, and who signs the written cure statement. Test it with a tabletop exercise.
- Review biometric and processor contracts: confirm § 503.001 consent and retention, and add AI-specific assistance clauses for processors under § 541.104.
Rollback: keep the previous model or agent version deployable and keep feature flags for AI functions, so a disputed feature can be withdrawn during the cure window without a full outage.
Framework mapping and measurement
| TRAIGA element | NIST CSF 2.0 | NIST SP 800-53 Rev. 5 | Other |
|---|---|---|---|
| Applicability and legal tracking | GV.OC-03 | PM-28 Risk Framing | ISO/IEC 27001:2022 A.5.31 |
| AI system inventory | ID.AM-07 | CM-8 System Component Inventory | NIST AI RMF Map function |
| Disclosure notices | GV.PO-01 | PT-5 Privacy Notice | OWASP LLM09 Misinformation |
| Agent permissions and intent | GV.PO-01 | AC-6 Least Privilege; IA-9 Service Identification and Authentication | OWASP LLM06 Excessive Agency |
| Post-deployment monitoring | DE.CM-09 | AU-2 Event Logging | NIST AI 600-1 |
| Cure statement and notice | RS.CO-02 | IR-6 Incident Reporting | Internal playbook |
Metrics that show the programme is working:
- Percentage of Texas-facing AI systems with a complete evidence pack (target 100 per cent).
- Percentage of AI agents with a named owner and a dedicated workload identity.
- Red-team findings closed within the agreed time, per release.
- Time to assemble the CID response set in a tabletop test (target well under the 60-day cure window).
- Health care and public-sector AI interactions shown with a disclosure, sampled monthly.
Common failure patterns
- Assuming "intent based" means "no work": without documented purpose and testing, intent is inferred from what the system did.
- Treating it as public sector only: the manipulation, discrimination, constitutional-rights and sexual-content bans bind any person.
- No owner for vendor AI: embedded AI features in SaaS tools are deployed by you, and nobody holds the evidence.
- Shared credentials behind agents: logs that show a generic service account cannot prove which system did what during a cure.
- Testing without records: red teaming that is not written down cannot support the § 552.105 defence.
Practitioner checklist
- Confirm in writing whether each AI system serves Texas residents.
- Record intended purpose and prohibited-use screening for every system and agent.
- Implement AI disclosures for health care and public-sector interactions.
- Assemble a per-system evidence pack matching the § 552.103 CID list.
- Map AI governance to NIST AI 600-1 and keep the mapping current.
- Schedule and document red-team testing before release and after major changes.
- Give every AI agent a named owner and its own scoped identity.
- Write and rehearse a 60-day cure playbook, including model rollback.
- Review biometric data handling against the amended § 503.001.
Standards and references
- Texas Legislature: HB 149 (89R), enrolled text
- Office of the Texas Attorney General: Texas Responsible Artificial Intelligence Governance Act
- NIST: AI 600-1, AI Risk Management Framework: Generative Artificial Intelligence Profile
- The White House: Fact Sheet, Ensuring a National Policy Framework for Artificial Intelligence
- NIST: Cybersecurity Framework 2.0
- NIST: SP 800-53 Rev. 5
- OWASP: Top 10 for LLM Applications 2025
This guide summarises the Texas Responsible Artificial Intelligence Governance Act as enrolled and its status as at 24 October 2026, and is not legal advice. Related NHI Mgmt Group resources: Global AI Regulation Map · AI Governance Guide · NIST AI RMF Guide · US State Privacy Laws Guide · Colorado AI Act Guide