
Short answer: yes, the Texas AI law is in force. The Texas Responsible Artificial Intelligence Governance Act, called TRAIGA, is House Bill (H.B.) 149 of the 89th Legislature (see the bill page). It took effect on January 1, 2026. It is mostly a list of banned uses, such as building AI to push people toward self-harm or to unlawfully discriminate on purpose. Only the Texas attorney general (the AG) can enforce it, apart from license sanctions (penalties on a professional or business license) by another state agency, which need a finding of violation and an AG recommendation first. A business gets 60 days to fix a problem after a written notice. This is not legal advice. Ask a lawyer whether it applies to you.
Reviewed and corrected on 12 October 2026. This is general information, not legal advice; check the cited statute or opinion before relying on it.
Official text, as of 12 October 2026
Texas Business & Commerce Code § 552.056(b), added by H.B. 149 (Acts 2025, 89th Leg., R.S., Ch. 1174): "A person may not develop or deploy an artificial intelligence system with the intent to unlawfully discriminate against a protected class in violation of state or federal law." Section 552.101(a) adds that "the attorney general has exclusive authority to enforce this chapter," except for agency license sanctions under § 552.106. Read the enrolled act at capitol.texas.gov and the codified chapter on the Texas statutes site.
What changed in 2026: the whole subtitle (chapters 551 to 554) and the amendments to § 503.001 took effect on January 1, 2026. The AG's deadline to post the online complaint mechanism was September 1, 2026.
TL;DR
- Status on October 12, 2026: in force since January 1, 2026. In our October 2026 review the Texas statutes site, which says it is current through the 89th Legislature's 2025 second called session, showed no amendment to this law, and we found no report of a court order stopping it.
- Who: anyone who does business in Texas, makes a product Texans use, or builds or deploys AI in Texas.
- What: bans on intentional misuse for everyone. Disclosure and social scoring rules for government. Disclosure rules for health care providers.
- Enforcement: the AG only, with a 60-day cure window (time to fix a problem before a lawsuit). The statute says it creates no private lawsuit.
- Versus Colorado: Texas asks businesses to avoid listed bad intent. It sets no duty for private businesses to write an impact assessment (a written risk review) and no general customer notice duty. The exceptions are the health care notice and the biometric consent rule.
The dates, in order
On October 12, 2026, we read the enrolled act (the final text both chambers passed), the legislature's bill page and the codified text (the act as filed in the state's code).
- March 14, 2025: H.B. 149 filed.
- April 23, 2025: passed the House.
- May 23, 2025: passed the Senate with amendments.
- May 30, 2025: the House agreed to the Senate changes.
- June 22, 2025: signed by the governor.
- January 1, 2026: the effective date, set by section 10 of the act.
- September 1, 2026: the deadline in section 8 for the AG to post an online complaint mechanism. The AG's Consumer AI Rights page carries a "File An AI Complaint" link.
The legislature's bill page lists these steps. The act became Subtitle D of Title 11 ("Personal Identity Information") of the Texas Business and Commerce Code, chapters 551 to 554, titled "Artificial Intelligence Protection." A subtitle is a group of chapters in a code. The codified text of chapter 554 ends with history lines such as "Added by Acts 2025, 89th Leg., R.S., Ch. 1174 (H.B. 149), Sec. 4, eff. January 1, 2026," and the amended section 503.001 ends with a matching line for Sec. 2 of the act. That means the 89th Legislature, Regular Session, chapter 1174. No later amendment is listed on either page. For why a law exists in two forms, see what a session law is.
Who it applies to
Section 551.002 says the subtitle "applies only to a person who" does one of three things. The person promotes, advertises or conducts business in Texas. Or the person produces a product or service used by Texas residents. Or the person develops or deploys an AI system in Texas. That reaches a company outside Texas whose product Texans use.
Section 551.001 defines an "artificial intelligence system" as a "machine-based system" that "infers from the inputs the system receives how to generate outputs." Outputs include "content, decisions, predictions, or recommendations."
A developer builds a system that is sold or offered in Texas. A deployer puts one to use in Texas. Section 552.001 defines both. A consumer is a Texas resident "acting only in an individual or household context." An employee at work is not a consumer under this law.
A "governmental entity" (section 552.001(3)) means a department, board, office or other administrative unit of the state or of a political subdivision that exercises governmental functions. It does not include hospital districts created under the Health and Safety Code or Article IX of the Texas Constitution, or an institution of higher education as the Education Code defines it, including a university system. The consumer disclosure rule in section 552.051(b) says "governmental agency" instead, a term the act does not define.
What it bans and requires
Most readers are surprised by how little it asks of a private business. The text has no impact assessment duty. It has no general customer notice duty.
| Duty | Who it applies to | Where in the statute |
|---|---|---|
| Do not build or use AI that "intentionally aims to incite or encourage" self-harm, harm to others or crime | Any person | § 552.052 |
| Do not use AI to impair rights under the U.S. Constitution, if that is the "sole intent" | Any person | § 552.055 |
| Do not build or use AI "with the intent to unlawfully discriminate against a protected class" (a group the civil rights laws protect, such as race, sex, age, religion or disability) | Any person, with exceptions for insurers and banks | § 552.056 |
| Do not develop or distribute AI with the "sole intent" of producing child pornography (Penal Code § 43.26) or unlawful deepfakes (fake images or video made with AI, Penal Code § 21.165). Do not intentionally develop or distribute AI that holds sexual text chats while imitating a child under 18 | Any person | § 552.057 |
| Tell people they are dealing with AI, before or at the time of use | A governmental agency offering a consumer-facing system | § 552.051(b) |
| Tell patients AI is used in their care, no later than the date the service is first provided (in an emergency, as soon as reasonably possible) | A health care provider | § 552.051(f) |
| No social scoring that leads to detrimental treatment in an unrelated setting, unjustified or disproportionate treatment, or infringement of legal rights | Government only | § 552.053 |
| No identifying people by biometric data (fingerprints, voiceprints, iris scans) or gathering their images from the Internet without consent, if it infringes rights. Photographs are not "biometric data" under this section | Government only | § 552.054 |
| Tell people and get their consent before capturing a biometric identifier (a retina or iris scan, fingerprint, voiceprint, or record of hand or face geometry) for a commercial purpose | Any person | § 503.001 |
Two points in the table need care.
Intent is the test. Section 552.056(c) says "a disparate impact is not sufficient by itself to demonstrate an intent to discriminate." A disparate impact is a bad result that falls harder on one group. A biased result alone does not prove a violation.
The notice rule is narrow. Section 552.051(b) says a "governmental agency that makes available an artificial intelligence system intended to interact with consumers shall disclose" this to each consumer "before or at the time of interaction." That is a duty for government. A private chatbot owner has no general duty under TRAIGA. The one private-sector notice is for health care, in section 552.051(f). When the duty applies, the disclosure must be clear and conspicuous, in plain language, and free of dark patterns (sections 552.051(c) to (e)), and it can be a hyperlink to a separate page.
Section 503.001 is an older biometric law that TRAIGA amended. It sits outside chapter 552, and the same AG can enforce it. A face image found online does not count as consent unless the person made it public (subsection (b-1)). Training, processing or storing biometric identifiers to develop AI models is excepted, unless the system is used to uniquely identify a specific individual (subsection (e)(2)), and a later commercial use brings the identifier back under the section (subsection (f)). A violation carries a civil penalty of up to $25,000 for each violation, and the AG may sue for it.
Limits and exemptions
- Section 552.056(d) exempts insurers already covered by insurance anti-bias laws from the discrimination ban.
- Under section 552.056(e), a federally insured bank "is considered to be in compliance" if it follows banking law.
- Section 552.002 says the chapter cannot be read to limit free speech. It also keeps insurance oversight with the Department of Insurance.
- The AG cannot seek a penalty over an AI system "that has not been deployed" (section 552.105(f)).
- Section 552.003 overrides any city or county rule on AI. That is called preemption, meaning the state law blocks local rules.
Who enforces it, and what it costs
Section 552.101(a) gives the AG "exclusive authority to enforce this chapter," except for licensing sanctions by other agencies. Section 552.101(b) adds: "This chapter does not provide a basis for, and is not subject to, a private right of action for a violation of this chapter or any other law." A private right of action is a person's right to sue directly.
A case moves in these steps. A civil penalty is a money penalty paid to the state.
- The AG must keep an online mechanism where a consumer can submit a complaint (§ 552.102).
- If a complaint comes in through that mechanism, the AG may issue a civil investigative demand, a written order to hand over information (§ 552.103). It can ask for the system's purpose, training data types, outputs, metrics, known limits and safeguards. The text does not say the demand is the only way the AG can learn of a violation.
- If the AG finds a violation, it must send written notice naming the provisions (§ 552.104(a)).
- The AG may not sue before day 60. It also may not sue if, within 60 days, you cure the violation and send a written statement with proof and policy changes (§ 552.104(b)).
- If you do not cure, the penalties in § 552.105 apply.
| Kind of violation (a court decides if it is curable or uncurable, meaning fixable or not) | Civil penalty (§ 552.105(a)) |
|---|---|
| Curable, or breaking a promise made in a cure statement | $10,000 to $12,000 each |
| Uncurable | $80,000 to $200,000 each |
| Continuing | $2,000 to $40,000 per day |
The AG can also seek an injunction, which is a court order to stop the conduct, plus fees and costs. Section 552.106 lets a state agency sanction a person it licenses, registers or certifies for a violation of the ban list (Subchapter B). The sanctions can include suspension, probation or revocation of the license and a fine of up to $100,000. They need a finding of violation under § 552.105 and an AG recommendation first.
The law helps defendants too. Section 552.105(c) sets "a rebuttable presumption that a person used reasonable care." That means a court starts by assuming you were careful unless the state proves otherwise. Under subsection (e), you may not be found liable if someone else uses your system in a prohibited way. You also may not be found liable if you discover the violation through feedback from a developer, deployer or other person, through testing such as red-team testing (testing where staff try to break the system), by following state agency guidelines, or through an internal review process. The internal review route applies only if you substantially comply with the most recent NIST "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile" or another nationally or internationally recognized AI risk framework.
The test program and the council
Chapter 553 sets up a regulatory sandbox, which is a program that lets a firm test an AI system under state watch without the usual license. The Department of Information Resources runs it. Tests can last up to 36 months, and the department may extend a test for good cause (§ 553.053). The AG cannot bring charges over a waived rule during the test. The ban list in Subchapter B of chapter 552 cannot be waived (§ 553.051(e)). Applicants need approval from the department and any applicable agency and must file quarterly reports. The statute tells the department to set the application form by rule (§ 553.052(b)). In our October 2026 review we found no application form or rule for this chapter 553 sandbox. The department's Public Sector AI Sandbox page describes a different program, for government bodies, created by S.B. 1964.
Chapter 554 creates the Texas Artificial Intelligence Council, with seven members. It advises the legislature. Section 554.103 says the council may not "adopt rules or promulgate guidance that is binding for any entity."
How it differs from Colorado
Colorado's law starts January 1, 2027, and a stipulated federal court order, on its wording, currently bars the Colorado AG from starting enforcement. It covers tools that help decide jobs, loans, housing and similar matters. Under sections 6-1-1702 and 6-1-1704 of the Colorado code, developers share documents and deployers send notices and 30-day letters. Texas has no such duties for private businesses. Texas bans listed bad purposes and leans on proof of intent. Both put enforcement mainly with the AG. Texas gives every violator a 60-day cure window, while Colorado's cure notice applies only where the AG deems a cure possible and ends on January 1, 2030. Colorado says its act creates no new private right of action. Texas says its chapter is not a basis for a private action for a violation of the chapter "or any other law." Texas lists penalty amounts. Colorado's part 17 lists none and sends enforcement through its Consumer Protection Act.
Do not mix it up with S.B. 1964
A separate 2025 law, Senate Bill 1964 (Chapter 1148, Acts 2025), added rules for government AI use in Government Code chapter 2054, effective September 1, 2025. Section 2054.708 requires a state agency, or a vendor that contracts with a state agency, to assess a "heightened scrutiny" system, meaning one "specifically intended to autonomously make, or be a controlling factor in making, a consequential decision." The assessment covers risks of unlawful harm, system limitations and information governance. It is not part of TRAIGA, and it binds government and its vendors.
A worked example
This is a made-up example. A Texas telehealth clinic adds an AI assistant that helps with patient care. Section 552.051(f) says that when AI is used in relation to a health care service or treatment, the provider must tell the patient, no later than the date the service is first provided. Whether a tool that only answers general questions counts as "used in relation to" care is not defined in the statute, so the clinic's lawyer would have to make that call. The clinic treats it as covered and adds the notice to its intake form. Months later, a patient complains through the AG's web mechanism. The office sends a demand for the system's purpose, data types and safeguards. The clinic's test logs and review notes now matter. If the office decides there is a violation, it must send written notice, and the clinic has 60 days. It fixes the issue, files its written statement, and the office may not sue.
TRAIGA compliance checklist for a business
- Work out which role you hold: developer, deployer, health care provider, collector of faces or fingerprints, or none of these. Then list where your AI touches Texans, including products sold from out of state.
- Look for any use that could be read as pushing self-harm, crime or intentional bias. Write down your purpose for each system.
- If you are a health care provider, add the disclosure before the first service.
- If you capture faces or fingerprints, get clear consent first.
- Keep test and review records. They can support the discovery-based defenses in § 552.105(e).
- Pick a person to answer an AG notice inside 60 days.
What goes wrong
The first error is using a Colorado checklist. A Texas business that builds impact assessments is solving the wrong problem. The second is treating a vendor's chatbot notice as a TRAIGA duty. The third is ignoring the 60-day clock. Miss it and the chance to cure is gone.
Federal action has not changed this. Executive Order 14365, signed December 11, 2025, calls for a review of state AI laws, and the Federal Register prints it at volume 90, page 58499. Its text names Colorado's law and says nothing about Texas. The Federal Trade Commission's July 1, 2026 release on a proposed policy statement likewise names Colorado's act, not Texas's. An executive order is a presidential directive to federal agencies, and it cannot repeal a state statute.
How to check the current text
Open the statute on the legislature's site. Read the history line under each section. It lists every act that changed it. Check again after the legislature's next regular session. This guide is part of US Law Data: The Complete Guide. To cite Texas law, see how to find and cite Texas law.
Your startup sells a customer-support chatbot to retailers that serve Texas shoppers. A sales lead says TRAIGA makes every chatbot tell users it is an AI. What does the statute's disclosure rule actually cover?
FAQ
Is the Texas AI law in effect? Yes. TRAIGA took effect on January 1, 2026. In our October 2026 review we found no amendment to it.
What does TRAIGA actually prohibit? It bans building or using AI with a bad aim. The listed aims are pushing people toward self-harm, harm to others or crime, impairing rights under the U.S. Constitution (sole intent), unlawful discrimination, and, with sole intent, producing child pornography or unlawful deepfakes. It also bans intentionally building AI that holds sexual text chats while imitating a child. Government also may not use social scoring, or identify people by biometric data without consent.
Does my business have to tell customers they are talking to AI? Not under TRAIGA, unless you are a governmental agency or a health care provider. Other laws may still require it.
Can I be sued by a customer under TRAIGA? The statute says it creates no private right of action. Only the AG can enforce it. Claims under other laws are separate.
What are the penalties? Civil penalties under section 552.105 run $10,000 to $12,000 for a curable violation and $80,000 to $200,000 for an uncurable one, plus $2,000 to $40,000 per day for a continuing one. The AG can also seek an injunction and fees, and a licensing agency can add sanctions.
Do I get a chance to fix a violation? Yes. The AG must give written notice and wait 60 days. If you cure and report in writing, no suit may follow.
Is there a Texas AI sandbox? The statute creates one, with tests of up to 36 months. In our October 2026 review we found no published application form or rule for it. The Department of Information Resources' sandbox page we found covers a different program for government bodies. Check dir.texas.gov for updates.
Where do I read the official text? The Texas statutes site has chapter 552. The legislature's bill page for H.B. 149 has the enrolled act.
For the US statutes, regulations and court rules these laws sit in, see the US primary law API.
New legal AI guides, weekly.
Further Reading
- AI Hiring and Employment Laws by State: What Is in Force
- AI Chatbot Disclosure Laws by State: When a Bot Must Say It Is AI
- California AI Laws Explained: SB 53 and the Rest
- New York RAISE Act Explained: Who It Covers and When It Starts
- Utah AI Law Explained From the Statute: What Is in Force Now
- Texas Attorney General Opinions: How to Find, Read and Cite Them

Co-Founder & CTO
Priyansh leads engineering and AI at Vaquill AI: the pipelines that pull statutes, regulations and court rules from every US jurisdiction's official publisher, and the REST API, MCP server and open dataset that serve them.
