Experimenting Under Guardrails: Texas's Sandbox, Council, and State Agencies
TRAIGA's regulatory sandbox is the clearest evidence that the Act is not only a catalogue of prohibitions. The question here is how Texas makes room for AI testing while retaining core protections, and how the Department of Information Resources, the Texas AI Council, applicable agencies, and the legislature fit around that experiment. Understanding those interfaces lets a reader distinguish a bounded innovation pathway from a general exemption and see how public-sector visibility becomes part of the same governance design.
The sandbox matters because regulation always has two jobs that can pull against each other. It must limit conduct that the legislature has treated as unacceptable, but it must also allow officials to learn about unfamiliar systems before ordinary rules fit them well. TRAIGA addresses that tension through conditional access, reports, oversight, and a line that testing cannot cross.
The sandbox is a bargain, not an exemption
The Act directs the Texas Department of Information Resources, in consultation with the Texas AI Council, to create a regulatory sandbox. The program is designed to give a person legal protection and limited access to the Texas market while testing an innovative AI system without obtaining a license, registration, or other regulatory authorization that would otherwise be required. Its stated sectors include health care, finance, education, and public services. It also contemplates research, training, testing, and other pre-deployment activity.
The exchange is straightforward in principle. A participant receives a controlled route to test a system; the state receives a defined description of the system, an account of its benefits and risks, a mitigation plan, and continuing information about performance and stakeholder experience. The program can waive or suspend certain laws and regulations during the testing period. In return, the participant accepts limits on time, scope, approval, reporting, and continued operation.
That exchange is not a general license to experiment without legal consequences. The requirements of Chapter 552, Subchapter B—the part containing the duties and prohibitions on the use of AI—may not be waived. The Attorney General and a state agency may pursue charges or action against a program participant that violates those requirements even during the testing period. The protection applies to laws or regulations actually waived under the program, not to the Act's core AI conduct boundaries.
The application process makes the bargain concrete. A person must obtain approval from the Department of Information Resources and from each applicable agency before testing under the program. An applicable agency is the state department that would regulate the relevant business activity if the participant were not operating within the sandbox. The participant must provide a detailed description of the system and its intended use, include a benefit assessment addressing potential effects on consumers, privacy, and public safety, describe how adverse consequences will be mitigated, and provide proof of compliance with applicable federal AI laws and regulations.
These requirements do more than create an intake form. They force the proposed experiment to become legible before it reaches the market.