Texas TRAIGA and the NIST AI RMF: the liability defense explained

A curable violation is one the person can fix after written notice from the Texas attorney general. Under Section 552.105, civil penalties for uncured curable violations range from $10,000 to $12,000 per violation. The classification turns on whether the harm is reversible, whether the conduct can be stopped and corrected, and whether the underlying practice can be changed without lasting consequences for affected people.

• Notice first: The attorney general must give written notice of the violation and may not sue before the 60th day after that notice under Section 552.104.
• Cure window: A person who cures the violation and provides a written statement with supporting documentation and the internal policy changes made to prevent further violations can avoid suit altogether.
• Documentation drives the outcome: Logs, change records, model updates and policy revisions are what show the violation was corrected.
• Reversible conduct: Output filtering, access restrictions or disabled features are typically easier to present as curable than harms that have already spread.
• Illustrative example: A deployed chatbot that produces a single prohibited output type, corrected by a documented filter update within the notice period, would be a candidate for this tier.

Confirm with counsel how the attorney general would classify your specific violation before you rely on a cure strategy.

An uncurable violation is one that cannot be undone once it has occurred, so the cure path under Section 552.104 may not be available. Section 552.105 sets civil penalties of $80,000 to $200,000 per uncurable violation. Counsel typically assesses whether the harm is complete, whether affected people cannot be made whole, and whether the conduct itself was the prohibited act.

• Completed harm: A violation whose consequences have already occurred, such as a deployed system that has already caused a prohibited result, is harder to treat as curable.
• Intent-based prohibitions: Practices banned for intent, including developing or deploying AI intended to incite harm or to unlawfully discriminate under Section 552.056, may be harder to cure because the intent itself is the violation.
• Limited cure record: Without a documented correction, the attorney general has less basis to treat the matter as curable.
• Scale of exposure: Each violation is assessed separately, so the number of affected people or outputs affects the total the reader should model.
• Ask for a quote on scope: Any remediation or counsel engagement should be scoped in writing so the response plan is ready before notice arrives.

Confirm the classification with counsel, since the same facts can be argued either way.

A continued violation is one that persists after the violation has been identified or after the period in which it could be cured. Under Section 552.105, civil penalties for continued violations range from $2,000 to $40,000 per day. The daily measure means that delay in stopping conduct can increase exposure faster than the underlying issue itself.

• Daily accrual: Each day the violation continues can be counted separately, so the time between notice and corrective action matters.
• Notice does not pause the clock automatically: The cure protection in Section 552.104 depends on a documented cure, so the reader should not assume that waiting preserves options.
• Kill-switch capability: Being able to disable or roll back an AI system quickly shortens the period of continued conduct.
• Monitoring evidence: Post-deployment monitoring records show when a problem was detected and when it was stopped.
• Internal escalation: Clear escalation paths from detection to a decision to suspend the system reduce the gap.

Confirm with counsel how continued days would be counted and whether any part of the period could be treated as a cured interval.

The organization that develops or deploys the AI system owns the defense. Section 552.105(e) turns on whether the organization discovered a violation through feedback, adversarial or red-team testing, state agency guidelines or an internal review process while substantially complying with the most recent NIST Generative AI Profile or another recognized framework. Only the organization can make those findings, act on them and decide what changes its policies will carry.

• Accountable executive: Name a person who signs the AI risk decisions, the deployment go or no-go and any residual risk acceptance.
• Internal review process: Run and document the reviews that surface violations, including who reviews, how often and how findings reach decision makers.
• Policy changes: Approve the internal policy changes that prevent further violations, since the cure statement must describe them.
• Cure statement: Sign the written statement that the attorney general requires under Section 552.104, and confirm every factual claim in it.

Before relying on any step, ask counsel which internal records the attorney general could request under Section 552.103 and confirm that your team can produce them.

An implementation partner such as DeployOpen prepares the systems, documentation and testing evidence that the organization relies on. The partner works inside the client's own infrastructure, so the logs, test results and configuration records stay under the client's control. The partner does not decide whether the organization has substantially complied, and it does not issue audit reports, attestations or certificates.

• Systems and controls: Build or harden the self-hosted platform, access controls and monitoring that the organization's program describes.
• Testing evidence: Run security testing of applications, APIs and AI systems, and document the methods, results and remediation.
• Documentation: Prepare system descriptions, deployment context records and post-deployment monitoring procedures that map to the NIST Generative AI Profile.
• Cure package inputs: Assemble the technical material the organization may need for a cure statement, such as change records and the evidence behind each policy update.

Confirm in the written scope which artifacts the partner delivers, who keeps custody of them and who signs off on each one.

Counsel advises on how Texas law applies to a specific system, including the intent standards in Section 552.056 and the prohibited practices in Sections 552.052 to 552.054. An independent assessor, where one is engaged, reviews the evidence against the organization's own program. No assessment body certifies TRAIGA compliance, and no NIST program issues a conformity assessment for the AI RMF or its Generative AI Profile. Any outside review is an opinion on the record, not a determination by the state.

• Legal conditions: Counsel interprets whether the defense conditions are met, including the requirement that the organization discovered the violation through a listed route.
• Evidence review: An independent reviewer tests whether the documentation supports each claim in the cure statement.
• Scope of opinion: Ask for written limits on what the reviewer covers and whether the reviewer expresses an opinion on legal liability.
• Enforcement context: Confirm that the attorney general holds exclusive enforcement authority under Section 552.101 and that the chapter creates no private right of action.

Ask whether the reviewer has prior work on AI governance or on Texas law, and whether it will state in writing that it has issued no certificate of TRAIGA compliance.