# HB 585 - Introduced

By: Bhojani H.B. No. 585

A BILL TO BE ENTITLED

AN ACT

relating to the requirements for artificial intelligence operations of state agencies, establishing artificial intelligence divisions.

BE IT ENACTED BY THE LEGISLATURE OF THE STATE OF TEXAS:

### SECTION 1.

Chapter 2054, Government Code, is amended by adding Subchapter T to read as follows: <ins>SUBCHAPTER T. STATE AGENCY ARTIFICIAL INTELLIGENCE GOVERNANCE AND PUBLIC DISCLOSURE</ins>

<ins>Sec. 2054.751. DEFINITIONS. In this subchapter:</ins>

<ins>(1) "Artificial intelligence division" means the organizational unit established by a state agency under Section 2054.752.</ins>

<ins>(2) "Artificial intelligence risk officer" means the employee designated under Section 2054.753.</ins>

<ins>(3) "Material change" means a change to an artificial intelligence system, the state data used by the system, or the intended use of the system that may materially affect the system's accuracy, security, privacy risks, legal compliance, or effects on individuals.</ins>

<ins>(4) "Public artificial intelligence impact statement" means the public-facing statement required by Section 2054.758 and does not include an impact assessment prepared under Section 2054.708.</ins>

<ins>Sec. 2054.752. ARTIFICIAL INTELLIGENCE DIVISION REQUIRED. (a) Each state agency shall establish and maintain an artificial intelligence division within the agency to oversee the procurement, development, deployment, and use of artificial intelligence systems by the agency or on behalf of the agency.</ins>

<ins>(b) The division shall coordinate compliance with this subchapter and other applicable law concerning artificial intelligence systems, including applicable rules adopted by the department.</ins>

<ins>(c) The executive head of the agency shall designate the personnel and resources necessary to carry out the division's duties. The agency may assign existing employees to the division, and the division may consist of a single employee. This section does not require the creation of a new full-time equivalent position.</ins>

<ins>(d) A state agency that maintains an existing organizational unit performing the duties required by this subchapter may designate that unit as the agency's artificial intelligence division.</ins>

<ins>(e) The agency remains responsible for compliance with this subchapter regardless of whether a system is furnished or operated by a contractor.</ins>

<ins>Sec. 2054.753. ARTIFICIAL INTELLIGENCE RISK OFFICER. (a) Each state agency shall designate an employee of the artificial intelligence division to serve as the agency's artificial intelligence risk officer. The agency may assign this duty to an employee who performs other duties.</ins>

<ins>(b) The artificial intelligence risk officer shall:</ins>

<ins>(1) promote the ethical procurement, development, deployment, and use of artificial intelligence systems, consistent with the agency's code of ethics and the Artificial Intelligence Risk Management Framework published by the National Institute of Standards and Technology;</ins>

<ins>(2) oversee the inventory and ongoing evaluation of agency artificial intelligence systems;</ins>

<ins>(3) ensure the completion of risk assessments and impact assessments required for heightened scrutiny artificial intelligence systems;</ins>

<ins>(4) review risk assessments and approve or deny deployment of heightened scrutiny artificial intelligence systems as provided by Section 2054.756; and</ins>

<ins>(5) coordinate employee training, incident response, vendor compliance, record retention, and preparation of public artificial intelligence impact statements.</ins>

<ins>(c) The artificial intelligence risk officer shall have direct access to the agency's executive head or the executive head's designee regarding material artificial intelligence risks and compliance deficiencies.</ins>

<ins>Sec. 2054.754. AGENCY ARTIFICIAL INTELLIGENCE CODE OF ETHICS. (a) Each state agency shall adopt and maintain a written artificial intelligence code of ethics that incorporates and is no less protective than the code of ethics established by the department under this chapter.</ins>

<ins>(b) The agency's code of ethics must establish policies and procedures addressing:</ins>

<ins>(1) human oversight and control, including appropriate review of system inputs and outputs and the ability to pause, restrict, or disable a system that produces harmful or inaccurate results;</ins>

<ins>(2) fairness, accuracy, and protection of legally protected rights, including data governance practices and monitoring throughout the system's life cycle;</ins>

<ins>(3) transparency, including disclosure of public-facing artificial intelligence interactions and consequential uses and a prohibition on representing an artificial intelligence system as a human being;</ins>

<ins>(4) data privacy and security, including data minimization, authorized uses of personally identifying information, access controls, security monitoring, and records retention;</ins>

<ins>(5) public and internal redress, including a point of contact for complaints and procedures to identify and remedy adverse effects of artificial intelligence systems;</ins>

<ins>(6) accountability for agency and contractor conduct, including appropriate contract terms and timely disclosure by vendors of known vulnerabilities and available remedies;</ins>

<ins>(7) employee training on lawful, accurate, safe, and ethical use of artificial intelligence systems; and</ins>

<ins>(8) regular evaluations of artificial intelligence systems and documentation of identified deficiencies and improvements.</ins>

<ins>(c) Each state agency shall make its code of ethics available to agency employees and publish the code on its public Internet website. The code must be reviewed at least annually and revised as necessary to remain consistent with department rules and applicable law.</ins>

<ins>Sec. 2054.755. INVENTORY, MONITORING, AND RECORDS. (a) Each state agency shall maintain an internal inventory of artificial intelligence systems the agency procures, develops, deploys, or uses, including systems used by a contractor on the agency's behalf. The inventory must identify each system's intended purpose, responsible agency unit, and whether the system is public-facing or subject to heightened scrutiny.</ins>

<ins>(b) The agency shall establish methods for regularly evaluating each deployed artificial intelligence system for accuracy, performance, safety, privacy, and continued compliance with the agency's code of ethics. An evaluation must occur at least annually and after a material change affecting a heightened scrutiny artificial intelligence system.</ins>

<ins>(c) The agency shall document evaluations, corrective measures, system modifications, and relevant complaints and shall retain records, including system inputs and outputs when applicable, in accordance with state records retention requirements and other applicable law.</ins>

<ins>Sec. 2054.756. RISK ASSESSMENTS FOR HEIGHTENED SCRUTINY ARTIFICIAL INTELLIGENCE SYSTEMS. (a) Before a state agency develops, procures, deploys, or uses a heightened scrutiny artificial intelligence system, and when a material change is made to the system, the agency shall complete a written risk assessment evaluating the probability and severity of potential harm.</ins>

<ins>(b) The risk assessment must consider and document:</ins>

<ins>(1) known security risks and available measures to mitigate those risks;</ins>

<ins>(2) performance measures, including accuracy, relevance, and operational efficiency;</ins>

<ins>(3) transparency measures, including available information about the system's decision-making methods, training data, and inputs and outputs used to monitor its decisions;</ins>

<ins>(4) the adequacy of human oversight, privacy protections, and data governance; and</ins>

<ins>(5) any additional safeguards necessary to address reasonably foreseeable risks.</ins>

<ins>(c) Before deployment, the artificial intelligence risk officer shall review the assessment and approve or deny deployment based on the documented risks and mitigation measures. The officer shall notify the executive head of the agency or the executive head's designee of a decision to approve deployment. An agency may require additional executive approval but may not dispense with the officer's review.</ins>

<ins>(d) A state agency shall retain each risk assessment and related records for the period required by the applicable records retention schedule.</ins>

<ins>Sec. 2054.757. IMPACT ASSESSMENTS FOR HEIGHTENED SCRUTINY ARTIFICIAL INTELLIGENCE SYSTEMS. (a) Before deploying or using a heightened scrutiny artificial intelligence system, and when a material change is made to the system, the state data used by the system, or the intended use of the system, a state agency shall conduct or obtain an impact assessment consistent with Section 2054.708.</ins>

<ins>(b) The impact assessment must include:</ins>

<ins>(1) a description of the system, including its model, available information regarding training data, and intended use;</ins>

<ins>(2) a description of the agency's planned use and identification of the employees responsible for deployment, monitoring, and evaluation;</ins>

<ins>(3) whether the system will process or store personally identifying information and, if so, whether that information may be used to train the system;</ins>

<ins>(4) reasonably foreseeable risks of harm in violation of state or federal law, including unlawful discrimination, and measures to mitigate those risks;</ins>

<ins>(5) known system limitations;</ins>

<ins>(6) procedures and intervals for monitoring outputs for accuracy and harmful effects; and</ins>

<ins>(7) the retention period for inputs and outputs and the method for deleting records when authorized by law.</ins>

<ins>(c) At the request of an agency, a vendor may prepare an impact assessment on behalf of the agency. The agency shall obtain and review the completed assessment and retain it as an agency record. The agency shall provide the assessment to the department on request, subject to Section 2054.708.</ins>

<ins>(d) An impact assessment under this section is subject to the confidentiality protections of Section 2054.708. An agency may not publish the confidential assessment as a substitute for the public statement required by Section 2054.758.</ins>

<ins>Sec. 2054.758. PUBLIC ARTIFICIAL INTELLIGENCE IMPACT STATEMENT. (a) Each state agency shall prepare, maintain, and publish a public artificial intelligence impact statement describing the agency's procurement, development, deployment, and use of artificial intelligence systems.</ins>

<ins>(b) The public artificial intelligence impact statement must contain, in plain language:</ins>

<ins>(1) a description of the agency's artificial intelligence systems, or groups of substantially similar systems used solely for internal administrative functions, and their general purposes;</ins>

<ins>(2) identification of each system that is public-facing or is a heightened scrutiny artificial intelligence system, including the public services or consequential decisions the system supports;</ins>

<ins>(3) a summary, for each system described by Subdivision (2), of material known limitations and reasonably foreseeable risks involving accuracy, fairness, privacy, security, and legally protected rights;</ins>

<ins>(4) a summary of safeguards, including human oversight, accuracy monitoring, data protection measures, and mitigation of identified risks;</ins>

<ins>(5) the identity of the agency unit responsible for the system, the date the statement was last updated, and the general schedule for future review; and</ins>

<ins>(6) instructions for obtaining information or submitting a complaint, including a publicly accessible method to contact the agency's artificial intelligence risk officer or a designated representative.</ins>

<ins>(c) A state agency shall publish the statement in a clear and conspicuous manner on the agency's primary public Internet website. The agency shall:</ins>

<ins>(1) provide a readily identifiable link labeled "Artificial Intelligence Impact Statement" on the agency's homepage in the primary navigation or another prominently displayed location;</ins>

<ins>(2) ensure that the statement is available without a fee, account registration, or password and is readily accessible to individuals with disabilities; and</ins>

<ins>(3) present the statement in clear, understandable language and not solely within a privacy policy, terms of use, or other unrelated document.</ins>

<ins>(d) The statement must be published before the agency initially deploys a public-facing or heightened scrutiny artificial intelligence system. The agency shall update the statement not later than the 30th day after a material change to a disclosed system or the deployment of another artificial intelligence system and shall review the statement at least annually.</ins>

<ins>(e) A state agency that does not deploy or use an artificial intelligence system shall publish a statement to that effect and shall review the statement at least annually.</ins>

<ins>(f) A public artificial intelligence impact statement under this section is a separate public disclosure and is not an impact assessment under Section 2054.708. An agency shall not include in the public statement an impact assessment made confidential by that section, personally identifying information, information made confidential by other law, information that would materially compromise cybersecurity, or information protected as a trade secret under applicable law. The agency shall provide a nonconfidential, general description of a withheld risk or safeguard when the description can be made without revealing protected information.</ins>

<ins>Sec. 2054.759. ACCEPTABLE USE POLICIES; TRAINING; VENDORS. (a) Before deploying or using a heightened scrutiny artificial intelligence system, each state agency shall identify approved uses and system limitations and adopt a written acceptable use policy that prohibits unapproved uses.</ins>

<ins>(b) Each agency shall provide employees and contractors who access, use, or manage a heightened scrutiny artificial intelligence system with training on the policy, material known risks, and appropriate mitigation practices before authorizing such access or use.</ins>

<ins>(c) A contract under which a vendor deploys or operates a heightened scrutiny artificial intelligence system for a state agency must require the vendor to implement the Artificial Intelligence Risk Management Framework published by the National Institute of Standards and Technology or a comparable risk management framework, disclose known material vulnerabilities and available remedies in a timely manner, and cooperate with required risk and impact assessments.</ins>

<ins>Sec. 2054.760. DEPARTMENT GUIDANCE; COMPLIANCE. (a) The department shall develop and make available a model public artificial intelligence impact statement and guidance for implementing this subchapter. The department may adopt rules necessary to implement this subchapter, including rules concerning accessible Internet publication and standardized disclosures.</ins>

<ins>(b) Not later than September 30 of each year, each state agency shall certify to the department that the agency maintains the division, risk officer, code of ethics, assessments, and published statement required by this subchapter. The certification must identify the Internet address at which the agency's statement is published.</ins>

<ins>(c) If the department determines that a state agency has not complied with this subchapter, the department shall notify the agency's executive head and provide a reasonable opportunity to correct the deficiency. The department may publish a list of agencies that have not corrected identified deficiencies after notice.</ins>

<ins>Sec. 2054.761. CONSTRUCTION. (a) This subchapter is cumulative of, and does not limit, duties imposed by another provision of this chapter or by rules adopted under this chapter.</ins>

<ins>(b) This subchapter does not require disclosure of information that is confidential or excepted from required disclosure under Chapter 552 or other law and does not modify the confidentiality of an impact assessment under Section 2054.708.</ins>

<ins>(c) This subchapter does not modify the obligations of local governments under other law or create a private cause of action.</ins>

### <ins>SECTION 2.</ins>

<ins>(a) Not later than March 1, 2028, each state agency shall establish the artificial intelligence division, designate the artificial intelligence risk officer, and adopt the code of ethics required by Subchapter T, Chapter 2054, Government Code, as added by this Act.</ins>

<ins>(b) Not later than September 1, 2028, each state agency shall publish its initial public artificial intelligence impact statement under Section 2054.758, Government Code, as added by this Act, covering systems in operation on that date. For a public-facing or heightened scrutiny artificial intelligence system first deployed on or after September 1, 2028, the agency shall publish the required information before deployment.</ins>

<ins>(c) Not later than March 1, 2028, the Department of Information Resources shall make available the model statement and guidance required by Section 2054.760, Government Code, as added by this Act.</ins>

<ins>(d) This section does not delay or suspend any requirement concerning an artificial intelligence system that applied to a state agency before the effective date of this Act.</ins>

### SECTION 3.

This Act takes effect September 1, 2027.
