Bill would audit AI use for bias

Summary#

HB 1032, titled “Regards AI use by state agencies and in labor and employment,” would create new sections of Ohio law and amend existing sections on civil rights, layoffs, public employees, and unemployment benefits. Its stated goal is to address the use of artificial intelligence by state agencies and in labor and employment; it is an introduced bill, not law. The supplied text does not include the full bill or the current versions of the laws it would amend, so the changes could not be fully checked against current law.

• State agencies would have to report their AI systems to the Ohio Department of Administrative Services (DAS). The DAS director would publish a yearly public inventory and assess systems for unlawful discrimination and disparate impact.
• State agencies generally could not buy AI unless its use is specifically authorized by law. They also could not use it for public-assistance functions or functions that materially affect people’s rights, civil liberties, safety, or welfare unless the use is specifically authorized by law. Before deploying AI they are authorized to buy, agencies would have to obtain an independent bias audit and publish its results at least 60 days before deployment.
• Employers would be barred from using automated employment decision processes in a way that causes certain adverse employment actions because of listed characteristics. They would also have to give advance written notice about use of the process, its trade name, and the types and sources of personal information it will process or analyze.
• Employers required to give notice of a covered mass layoff or plant closing would have to report whether it is related to AI or another technological change. Unemployment applications and certain employer reports would also ask whether a layoff was due to AI.
• Use of AI by a public employer would be expressly included among matters subject to collective bargaining, with existing management rights also addressed in the bill.

What it means for you#

• State agency users and taxpayers: Agencies would have to inventory their AI systems and report details including uses, vendor, access to personal information, risks, and the cost of time savings. The director would have to make the inventory public and assess systems for discrimination. The bill does not set out a process for individuals to challenge an agency’s use of AI.
• People applying for or working in jobs: Employers would have to give advance written notice if an automated employment-related decision process will be used in a decision about employment. The bill’s definition includes, for example, resume screening and analysis of online interview speech or facial expressions. An individual could file an employment discrimination charge over prohibited use or failure to give notice.
• Workers affected by layoffs: If a covered layoff or plant closing is related to AI or another technological change, employers would have to say so in notices to state and local officials. The bill also adds AI-related questions to some unemployment paperwork; it does not change who qualifies for benefits in the text provided.
• Public employees and unions: The bill would expressly make a public employer’s use of AI a subject of collective bargaining. It does not remove the public employer’s stated management rights, while specifying that employment effects, including AI use, are subject to bargaining.
• People receiving public assistance: The bill would restrict agency use of AI in functions related to delivering public assistance benefits unless the use is specifically authorized by law. It does not itself change benefit amounts or eligibility rules.

Money#

No publicly available information.

• The bill would require agency staff time for inventories and assessments, and require independent audits before deployment of AI that agencies are authorized to buy. It gives no cost estimate or staffing number.
• Employers could face costs to give required notices and report layoff information. No estimate is provided.

What is unclear#

• The supplied text ends partway through the unemployment law amendments and does not include the full bill, including the proposed sections on bias audits. No publicly available information.
• The bill uses different definitions for AI in different sections. How those definitions would apply to particular systems is unclear.
• The bill does not specify the full audit or impact-assessment methods, how often assessments must occur, or what happens if an assessment finds a problem.
• The text provided does not state when the bill would take effect. Its status is introduced, so passage is uncertain.
• The parent laws were not supplied, so the changes could not be verified against current Ohio law.

Case for#

• A public inventory could help residents see which AI systems state agencies use, what they do, and what risks agencies identify.
• Independent bias audits before deployment and ongoing agency assessments could help detect discriminatory effects before or during use.
• Advance notice would tell job applicants and employees when automated processes are used and what personal information they analyze.
• Reporting whether layoffs are related to AI could give public officials and affected workers clearer information about the stated cause.

Case against#

• The restriction on buying AI, and on using it for specified public functions, is broad: agencies would need specific legal authorization. The bill does not explain how that authorization would be defined or how it would work with existing services.
• Audit, assessment, inventory, and notice duties could require staff and money, but the bill materials provide no estimates.
• The bill does not fully explain audit standards, oversight of the director’s assessments, or what remedies follow if an agency system is found to cause unlawful discrimination or disparate impact.
• Different definitions of AI and automated decision processes could make it unclear which systems are covered and what agencies or employers must do.