The Texas Responsible AI Governance Act (TRAIGA) is the Texas AI law that has applied since January 1, 2026 to anyone who develops or deploys an AI system in Texas, or whose products reach Texas residents. It has no high-risk category and no impact assessment. It prohibits a short list of intentional misuses, puts disclosure duties on government agencies and health care providers, and leaves enforcement to the Attorney General alone. What most companies actually need is the ability to answer an investigator’s questions about a deployed system, quickly and in writing.
Picture how a TRAIGA case would start. A Texas resident has an unsettling exchange with a company’s support chatbot, opens the Attorney General’s Consumer AI Rights page and clicks “File An AI Complaint”, which leads to the office’s general consumer complaint portal. That one form is enough to let the state ask the company for a description of the system, its training data, its metrics and its safeguards.
Texas Responsible AI Governance Act at a glance
TRAIGA is House Bill 149 of the 89th Legislature, which the governor signed on June 22, 2025. It adds chapters 551 to 554 to the Business and Commerce Code and took effect on January 1, 2026.
The act’s reach is broad. It applies to any person who promotes, advertises or conducts business in Texas, produces a product or service used by Texas residents, or develops or deploys an AI system in the state. The definition of an AI system is just as wide, covering any machine-based system that infers from its inputs how to generate outputs (content, decisions, predictions or recommendations) that can influence physical or virtual environments. An LLM behind a help widget plainly fits, and so, in our reading, does a trained scoring model inside a rules engine.
The duties themselves are narrow, though, and the word “consumer” narrows them further. In the statute it means a Texas resident acting in an individual or household context, and it explicitly excludes anyone acting in a commercial or employment context. That definition decides who can file through the complaint mechanism, and it is the first thing to check when someone tells you TRAIGA regulates your hiring tool.
What TRAIGA prohibits, and what it does not
Sections 552.052 to 552.057 hold six prohibitions, and four of them bind private companies. Every one of those four is written around intent.
- Manipulation. No one may develop or deploy an AI system “in a manner that intentionally aims to incite or encourage” a person to commit physical self-harm (including suicide), harm another person, or engage in criminal activity (section 552.052).
- Constitutional rights. No AI system may be built or deployed with the sole intent to infringe, restrict or impair a person’s rights under the US Constitution (552.055).
- Unlawful discrimination is barred only where a system is developed or deployed with the intent to discriminate against a protected class in violation of state or federal law (552.056). The statute adds that “a disparate impact is not sufficient by itself to demonstrate an intent to discriminate”. Regulated insurers are outside the section, and a federally insured financial institution that follows banking law is treated as compliant.
- Sexual content rules forbid building or distributing a system with the sole intent of producing child sexual abuse material or unlawful sexual deepfakes, or one that engages in sexual text conversations while impersonating a child under 18 (552.057).
Two further prohibitions apply only to governmental entities. Section 552.053 bans social scoring that leads to detrimental or disproportionate treatment, and 552.054 bars using biometric data, or images scraped from the internet, to uniquely identify a person without consent where that would infringe their rights. Section 2 of the bill also tightens the state’s biometric identifier law. A photo being public online no longer amounts to consent unless the person made it public, with exemptions for model training and for security and fraud work.
What the list leaves out matters as much, because plenty of TRAIGA guides describe duties the enrolled text does not contain. Some secondary summaries say the law restricts “high-risk” uses including certain applications in hiring. Our reading of the statute finds no high-risk category, no impact assessment, no duty to publish a risk program, and no hiring-specific obligation. Employment use is outside the “consumer” definition, so a discriminatory screening tool is reached only through the intent-based section 552.056, alongside existing employment law.
That narrowness cuts both ways. If you built a program for Colorado’s replacement AI law, most of it carries over and some of it goes beyond what Texas asks. If you built nothing, TRAIGA will not force a full program on you, but an investigative demand will still ask for the documents a program produces.
TRAIGA also preempts local AI ordinances inside Texas (section 552.003). Federally, an executive order signed on December 11, 2025 set up a Justice Department task force to challenge state AI laws. An executive order cannot invalidate a state law by itself, though, so TRAIGA stays enforceable unless a court or Congress says otherwise.
Who must disclose AI use under the Texas AI law
Disclosure reaches fewer organizations than people assume. Section 552.051 requires a governmental agency that makes an AI system available to interact with consumers to tell each person, before or at the time of the interaction, that they are dealing with AI. That applies even when it would be obvious. The notice must be clear, conspicuous and in plain language, may not rely on a dark pattern, and can be delivered through a hyperlink.
The second group is health care. When AI is used in relation to a health care service or treatment, the provider must give the same disclosure to the patient or their representative by the date the service is first provided. In an emergency it comes as soon as reasonably possible. Outside agencies and health care, section 552.051 imposes no disclosure duty, so a private retailer’s chatbot is not covered by it, although other consumer protection law still applies to what that chatbot says.
How TRAIGA enforcement works, from complaint to penalty
Enforcement belongs to the Attorney General exclusively, with no private right of action. Section 8 of the bill gave the office until September 1, 2026 to put its online complaint mechanism live, and the Cloud Security Alliance reports it was activated that day.
That brings us back to the chatbot complaint. Once a complaint arrives through the mechanism, section 552.103 lets the Attorney General issue a civil investigative demand, and the statute lists what it can ask for:
- A high-level description of the system’s purpose, intended use, deployment context and benefits.
- The type of data used to program or train it.
- The categories of data it processes as inputs, and the outputs it produces.
- Any metrics used to evaluate its performance, and its known limitations.
- A description of post-deployment monitoring and user safeguards, including, for deployers, the oversight, use and learning process for issues that arise.
- Any other documentation reasonably necessary for the investigation.
A violation finding comes as a written notice naming the provisions. No action can be filed before the 60th day. None can be filed at all if, within that window, the company cures the problem and sends a written statement covering the cure, the supporting documentation and the policy changes made to prevent a repeat. The 60 days run from the date the Attorney General provides the notice, not from when it reaches the right desk, so route it like a security incident.
Penalties apply only to violations that are not cured. Under section 552.105, a court can impose $10,000 to $12,000 for each curable violation (or for breaching the cure statement), $80,000 to $200,000 for each violation it finds uncurable, and $2,000 to $40,000 for each day a violation continues. For licensed professionals there is a second layer, because a state licensing agency may suspend or revoke a license and impose up to $100,000 once a violation has been found and the Attorney General recommends it.
The office cannot seek a penalty for a system that has not been deployed, and the statute sets a rebuttable presumption that the company used reasonable care. Per-violation counting is where exposure grows, which is why the cure window deserves a named owner.
The TRAIGA defence most teams can already earn
Section 552.105(e) is the part of the law security teams should care about most. A defendant may not be found liable if someone else used its system in a prohibited way, or if the company itself discovered the violation through one of four routes. The first three are feedback from a developer, deployer or other person, testing “including adversarial testing or red-team testing”, and following state agency guidelines. The fourth is an internal review process, which counts when the company substantially complies with the NIST AI RMF Generative AI Profile (NIST AI 600-1) or another nationally or internationally recognized AI risk framework.
It rewards discovery you can prove. A red-team exercise that found a jailbreak steering the support bot toward self-harm content, logged with a date, a ticket and a fix, is evidence. The same finding mentioned in a hallway is not. We think this is the most practical reason to formalize AI red teaming on Texas-facing systems, and to keep its findings in the same system of record as your vulnerability management.
What to do now: a TRAIGA compliance checklist
- Inventory every AI system a Texas resident can reach, including vendor models inside your products, and note whether you develop or deploy each.
- Test each against sections 552.052 to 552.057, writing down why each prohibition does or does not apply.
- Build the demand packet now, meaning the six items section 552.103 lists for each system, versioned and owned.
- Log adversarial testing and internal review against NIST AI 600-1, with findings and remediation dates, so the section 552.105(e) defence has a record behind it.
- Review marketing and product copy for claims that could read as intent, because intent is what the private-sector prohibitions turn on.
- Give legal a playbook for Attorney General notices and the 60-day cure clock, and confirm disclosure wording if you are a health care provider or agency.
If step 3 or 4 is where you stall, our AI governance and assurance work is built around exactly that evidence, and our ISO 42001 certification guide covers the management system that keeps it current.
Frequently asked questions
When did TRAIGA take effect?
The Texas Responsible AI Governance Act took effect on January 1, 2026. The governor signed HB 149 on June 22, 2025. The complaint mechanism that triggers a civil investigative demand had a deadline of September 1, 2026. Texas AI Report found no publicly reported TRAIGA enforcement action or demand through its September 29, 2026 check.
Does TRAIGA have a private right of action?
No. Section 552.101 gives the Attorney General exclusive enforcement authority and rules out private lawsuits under the chapter. Individuals take part by filing complaints through the Attorney General’s online mechanism, which serves Texas residents acting in a personal or household capacity.
Does TRAIGA apply to companies outside Texas?
Yes, if the company promotes, advertises or conducts business in Texas, produces a product or service Texas residents use, or develops or deploys an AI system in the state. A SaaS provider with Texas customers is in scope, although the prohibitions still require the intent the statute describes.
Is TRAIGA like the Colorado AI Act?
It is narrower. Colorado’s replacement law requires notices, explanations and human review for consequential decisions in defined domains. TRAIGA prohibits intentional misuse, requires disclosure only from government agencies and health care providers, and asks for documentation only when an investigation starts.
Bottom line
Assume the first TRAIGA question you receive will be a civil investigative demand triggered by one consumer complaint. Build the six-part demand packet for every Texas-facing AI system this quarter, and start logging red-team and internal review findings with dates and fixes, because that record is both your answer to the demand and your statutory defence.