Skip to main content

EPC Group's AI governance practice treats the Texas Responsible Artificial Intelligence Governance Act as an evidence problem. The Attorney General cannot sue until 60 days after written notice of a violation, and the law gives a defense to a company that discovers a violation through an internal review process while substantially complying with the NIST AI RMF Generative AI Profile or another recognized AI risk framework. EPC Group builds the supporting record from the Microsoft 365 tenant a company already runs.

Key Facts

  • In force: the Texas Responsible Artificial Intelligence Governance Act (TRAIGA, House Bill 149) was signed on June 22, 2025 and took effect on January 1, 2026.
  • Reach: it applies only to a person that promotes, advertises or conducts business in Texas, produces a product or service used by Texas residents, or develops or deploys an AI system in Texas.
  • Design: the AI-specific prohibitions that reach private companies turn on intent. HB 149 also amended Texas's biometric identifier law, which applies to private companies.
  • Enforcement: the Texas Attorney General enforces the law, and state licensing agencies can add sanctions against their own licensees only after a violation is established and the Attorney General recommends it. There is no private right of action.
  • Cure: the Attorney General cannot sue until 60 days after written notice, and cannot sue at all if the company cures the violation in that window and provides a written statement with supporting documentation.
  • Defense: a company is not liable if it discovers a violation through testing, feedback, state agency guidelines, or an internal review process while substantially complying with the NIST AI RMF Generative AI Profile or another recognized framework.
  • EPC Group's position: the record that answers a civil investigative demand or supports a cure statement (an inventory, a purpose statement and a monitoring record for each AI system) should exist before any notice arrives.

Last updated by Errin O'Connor, Founder & Chief AI Architect, EPC Group

By Errin O'Connor — Founder & Chief AI Architect, EPC Group · Microsoft Solutions Partner across all six designations · G2 Leader in BI consulting

What is TRAIGA, and who does it reach?

This page describes a statute for technology and compliance leaders. It is not legal advice. Confirm how the law applies to your organization with counsel.

TRAIGA is Texas's AI governance statute. It applies to developers, whose AI systems are offered, sold, leased or otherwise provided in Texas, and to deployers, who deploy AI systems for use there. The definition of an AI system is broad and technology-neutral. [5][1]

A company does not need a Texas headquarters to be covered. Producing a product or service used by Texas residents, or promoting, advertising or conducting business in Texas, is enough. [5] For most enterprises that run Microsoft 365 Copilot, Copilot Studio agents or Azure-hosted models, the likely role is deployer, and a company that builds its own agents may also be a developer. The statute does not define "deploy," so counsel should confirm the role.

What does TRAIGA prohibit for private companies?

The enacted text's prohibitions that apply to developers and deployers generally are these. [5]

  1. Developing or deploying an AI system intended to incite or encourage a person to commit self-harm, to harm others, or to engage in criminal activity.
  2. Developing or deploying an AI system with the sole intent that it infringe, restrict or otherwise impair an individual's rights under the U.S. Constitution.
  3. Developing or deploying an AI system with the intent to unlawfully discriminate against a protected class. Insurance entities and federally insured financial institutions have specific exemptions.
  4. Developing or distributing an AI system with the sole intent of producing or distributing child sexual abuse material or sexually explicit deepfake videos or images that violate the Texas Penal Code, or intentionally developing or distributing an AI system that engages in sexual text conversations while impersonating a child under 18.

The discrimination provision requires proof of intent. A disparate impact alone is not enough to establish a violation. [5]

That design matters for how a company prepares. An intent-based law is answered with a record of what each system was built to do, what it was tested for, and what the company did when something went wrong.

What applies only to government and health care?

Some duties bind only governmental entities: disclosing to consumers that they are interacting with AI, a ban on social scoring, and limits on using AI to identify people from biometric data or gathered images without consent. Separately, HB 149 amended Texas's existing biometric identifier law, which applies to private companies. Health care providers that use AI in a service or treatment must disclose it no later than the date the service is first provided, or as soon as reasonably possible in an emergency. [5]

Commentators note that the enacted bill is narrower for the private sector than the version first proposed. [2] A company that serves government or health care customers in Texas should expect those customers to pass their duties down in contracts.

How is TRAIGA enforced?

  • Who: the Texas Attorney General. There is no private right of action. Consumers can file complaints through an online mechanism the statute required the Attorney General to post by September 1, 2026. State licensing agencies can add sanctions against their own licensees, up to $100,000, after a violation is established and the Attorney General recommends it. [5][3]
  • Investigation: after a complaint through the online mechanism, the Attorney General can issue a civil investigative demand. The statute lets it request high-level descriptions of a system's purpose, intended use and deployment context, the type of data used to train it, its input data categories and outputs, performance metrics, known limitations, and post-deployment monitoring and user safeguards. [5]
  • Cure: the Attorney General cannot sue until 60 days after written notice. It cannot sue at all if, within that window, the company cures the violation and gives the Attorney General a written statement that it cured it, with supporting documentation and any necessary changes to internal policies. [5]
  • Penalties: under the statute, civil penalties run from $10,000 to $12,000 per violation that a court finds curable or per breach of a cure statement, from $80,000 to $200,000 per violation found uncurable, and from $2,000 to $40,000 for each day a violation continues. The statute also presumes, subject to rebuttal, that a person used reasonable care. [5]

Sixty days is generous if the record exists. It is not long enough to build an AI inventory from nothing.

What defenses does TRAIGA list?

TRAIGA says a defendant may not be found liable in these circumstances. [5]

  • Another person misused the AI system affiliated with the defendant.
  • The defendant discovered the violation through feedback from a developer, deployer or other person.
  • It discovered the violation through testing, including adversarial or red-team testing.
  • It discovered the violation by following guidelines set by applicable state agencies.
  • It discovered the violation through an internal review process while substantially complying with the most recent NIST AI RMF Generative AI Profile or another nationally or internationally recognized AI risk framework. [5][7][4]

Testing and internal review are things a company controls. Both are documentation exercises as much as technical ones: a test supports the defense only if the result was recorded, and framework alignment only if it can be shown.

The 60-Day Evidence Test

The 60-Day Evidence Test is EPC Group's readiness check for TRAIGA: for every AI system in the tenant, could the company produce the evidence below quickly enough to answer a civil investigative demand or support a cure statement? Rows 2 to 6 follow the items the statute lets the Attorney General request; rows 1 and 7 are EPC Group's additions. Each row names where the evidence lives in a Microsoft environment.

#Evidence itemWhere it lives in the Microsoft tenantWhat EPC Group checks
1An inventory of AI systemsMicrosoft Agent 365 registry; Copilot Studio environments; Azure AI Foundry projectsOne list, reconciled, with an owner per system
2Description and intended purposeThe agent register; solution documentationA written purpose statement, dated and approved
3Data the system readsMicrosoft Purview labels and DSPM for AI; SharePoint and OneDrive permissionsWhat each system can reach, and that labels apply
4Inputs and outputsPurview audit records of Copilot and agent interactionsAudit is on and retained for the period counsel sets
5Performance and testingFoundry evaluations; red-team and adversarial test recordsTests exist, are dated, and findings were closed
6Known limitations and safeguardsApproval gates, DLP policies, Conditional Access for agentsControls are configured, not only described
7Framework alignmentPurview Compliance Manager assessment for NIST AI RMF [8]An assessment with evidence attached, not a blank template

The check is met when all seven can be exported for every system.

Where do Copilot and agents create exposure?

TRAIGA does not ban Copilot or agents. The exposure is that a company cannot show what a system was intended to do.

  • Agents that touch people decisions. An agent that screens applicants, ranks employees or routes customer complaints is the kind of system whose purpose and testing should be documented first. The purpose statement and the test record are the answer.
  • Agents nobody owns. An agent built by a business user in Copilot Studio, with no registered owner and no stated purpose, has no one to speak to its intent. EPC Group's nine agent controls end in a register for that reason.
  • Overshared content. A system that can read everything can output anything. Oversharing remediation narrows what must be explained later.
  • Unsanctioned tools. AI used outside the tenant leaves no record at all. See shadow AI amnesty audit.

How does TAR-8 produce the evidence?

EPC Group's 8-surface Tenant AI Readiness Standard, TAR-8, scores a Microsoft environment on the eight surfaces an AI model reads, and its output is an evidence pack per surface. Surface 8, ownership and evidence, is the one TRAIGA leans on: named owners, one registry, an audit trail and a framework mapping.

TAR-8 is EPC Group's published standard. It is not a legal opinion, a Microsoft program or a certification, and passing it does not establish compliance with any statute. It produces the artifacts counsel will want in hand.

For the framework itself, see EPC Group's NIST AI RMF implementation guide and standards alignment.

A 30-day plan

  1. Days 1 to 7: inventory. Every Copilot surface, Copilot Studio agent, Foundry project and third-party AI tool, each with an owner.
  2. Days 8 to 14: purpose and data. A purpose statement per system; what each can read; labels and permissions checked.
  3. Days 15 to 21: testing and logging. Audit confirmed on; adversarial tests run on the systems that affect people; findings logged.
  4. Days 22 to 30: framework and rehearsal. NIST AI RMF assessment opened in Compliance Manager with evidence attached; a dry run of the 60-Day Evidence Test with counsel in the room.

Quick facts

QuestionAnswer
Is TRAIGA in force?Yes, since January 1, 2026
Who enforces it?The Texas Attorney General; state licensing agencies can add sanctions against licensees after a violation is established
Can individuals sue under it?No. Consumers can file complaints with the Attorney General
How long is the cure period?60 days from written notice, with a written cure statement
What defenses does it list?Third-party misuse, or discovering the violation through feedback, testing, state agency guidelines, or an internal review that substantially complies with the NIST AI RMF Generative AI Profile or another recognized framework
Is a company that only uses Copilot covered?Likely, as a deployer of an AI system for use in Texas; the statute does not define "deploy," so confirm with counsel
What does EPC Group produce?An inventory, a purpose statement and a monitoring record for each AI system, plus a framework assessment

What changed in 2026?

  • January 1: the law took effect. [6]
  • Complaint intake: the statute required the Attorney General to post an online complaint mechanism by September 1, 2026. [5] Check the Attorney General's site for its current form before relying on any description of it.
  • A registry became available. Microsoft Agent 365 reached general availability on May 1, 2026 and provides one registry of agents with governance actions. That is the first row of the evidence test for Microsoft-hosted agents. See Microsoft Agent 365 consulting.
  • Framework templates. Purview Compliance Manager carries assessment templates for NIST AI RMF and ISO/IEC 42001, which gives the framework-alignment record a place to live with evidence attached.
  • Metered AI. Usage-billed Copilot services arrived during the year, which means more AI systems in use than the license count suggests. See Copilot Cowork: access, cost and controls.

Who decides what?

DecisionOwnerEvidence it produces
Whether and how the statute appliesGeneral counselA written applicability memo
Which AI systems exist and who owns eachCIOThe inventory and register
What each system may read and doCISO with the data ownerLabel, permission and policy exports
What each system is forThe business owner of the systemA dated purpose statement
Which framework the company aligns toThe AI governance ownerThe Compliance Manager assessment
Who answers a civil investigative demandGeneral counsel, with a named technical leadThe response plan and its rehearsal record

The technical lead matters. Counsel will be asked for exports that only someone with the right admin roles can produce, on a clock.

How EPC Group delivers this

  1. The evidence test. EPC Group runs the 60-Day Evidence Test against the tenant and returns a pass or fail per system, per row, with the gap named.
  2. Remediation. EPC Group closes the gaps in order: inventory and owners first, then purpose statements, then data scope, then logging and testing.
  3. The standing record. EPC Group sets a quarterly review so the evidence stays current, run by the client's AI owner or by EPC Group's virtual Chief AI Officer.

The deliverables are an AI system inventory with owners, a purpose statement per system, a data-scope report, an audit and retention confirmation, a test log, a NIST AI RMF assessment with evidence, and a response plan for counsel. Engagements are fixed-fee and scoped after a call.

Terms defined

  • Developer: a person who develops an AI system that is offered, sold, leased, given or otherwise provided in Texas.
  • Deployer: a person who deploys an AI system for use in Texas.
  • Civil investigative demand: the Attorney General's written request for documents and information during an investigation.
  • Cure period: the 60 days after the Attorney General's written notice. If the company cures the violation in that window and provides a written statement with supporting documentation and any policy changes, the Attorney General cannot sue.
  • Statutory defense: a circumstance in which the statute says a defendant may not be found liable.
  • Red-teaming: deliberate adversarial testing of a system to find failures before someone else does.

Best for / Not the right fit for

Best for

  • Companies that do business in Texas and run Microsoft 365 Copilot, Copilot Studio agents or Azure AI.
  • General counsel and CIOs who need one evidence pack that serves the Attorney General, the board and the cyber insurer.
  • Organizations that want a named executive owner for AI through a virtual Chief AI Officer.

Not the right fit for

  • Organizations looking for a legal opinion. EPC Group is not a law firm.
  • Companies with no AI in production and no plan to deploy.

Where EPC Group fits

EPC Group has delivered Microsoft consulting since 1997 and holds all six Microsoft Solutions Partner designations. Its AI governance work includes more than 240 readiness assessments since August 2025. The practice is described at AI governance consulting; the firms EPC Group is compared with are listed in top AI governance consulting firms.

The data behind this (sources and verification)

Third-party figures above are attributed to their named sources as of the Last verified date. EPC Group audit figures are directional findings from client engagements.

Frequently asked questions

Does TRAIGA apply to private companies?

Yes. HB 149 applies to any person that promotes, advertises or conducts business in Texas, produces a product or service used by Texas residents, or develops or deploys an AI system in Texas. EPC Group advises clients to confirm their own position with counsel.

Does TRAIGA apply to Microsoft 365 Copilot?

HB 149 defines a deployer as a person who deploys an AI system for use in Texas, and a company that builds Copilot Studio agents may also be a developer. Whether either term fits is a question for counsel. EPC Group's work is to make sure each system has an owner, a stated purpose and a record.

What does TRAIGA prohibit?

For private companies, AI developed or deployed with intent to incite harm or crime, to infringe constitutional rights, or to unlawfully discriminate, and AI developed or distributed to produce certain illegal sexual content. EPC Group notes that each turns on intent.

Who enforces TRAIGA?

The Texas Attorney General. There is no private right of action, though state licensing agencies can add sanctions against their own licensees after a violation is established. EPC Group prepares clients for a civil investigative demand, which the Attorney General can issue after a complaint.

How long is the cure period?

Sixty days from the Attorney General's written notice, and a cure also needs a written statement with supporting documentation. EPC Group's 60-Day Evidence Test measures whether a company could assemble that record quickly.

Is following the NIST AI RMF a safe harbor under TRAIGA?

Not on its own. A company is not liable if it discovers a violation through an internal review process while substantially complying with the NIST AI RMF Generative AI Profile or another recognized framework. EPC Group records that alignment in Purview Compliance Manager with evidence attached.

Does TRAIGA require a specific product or a Microsoft license?

No. TRAIGA requires no product or license, and it does not require a private company to keep an AI inventory or adopt a framework. EPC Group assembles the records about a company's own systems that would answer a civil investigative demand.

How does TRAIGA relate to other state AI laws?

Other states have their own statutes with different duties and dates. EPC Group builds one inventory and one framework mapping so the same evidence serves each of them, and leaves the legal reading to counsel.

What does an engagement cost?

EPC Group engagements are fixed-fee and scoped after a call with a senior architect. EPC Group does not publish price ranges.

Ready to act on this?

Start with the practice most relevant to your estate, or reach out directly for a senior-architect conversation.

📧 contact@epcgroup.net · 📞 (888) 381-9725

Multiple models. One truth.

AI assistant — not a human