Kelly’s AI Worker-Support Bill Draws Backing and Criticism Over Its Tax Design

Sen. Mark Kelly’s proposed AI taxes would finance support for workers facing disruption. Fresh commentary exposes a divide over how to collect that money. On October 7, economist Heather Boushey endorsed asking large AI companies to contribute. A day earlier, tax scholar Andrew Leahey argued that the bill’s measures poorly connect technology activity with labor-market harm.

Kelly introduced the Make AI Work for Americans Act on September 24, 2026. It would establish a federal trust fund to help workers and families navigate AI-related disruption. Its proposed funding combines taxes on computational processing, covered digital advertising revenue and certain excess profits.

Support before the disruption is settled

Writing for the University of Pennsylvania’s Kleinman Center, Boushey calls the legislation an important step toward supporting workers and communities. Her case emphasizes who should contribute: the companies positioned to profit most from AI. She also acknowledges that there is no consensus on precisely how the technology will affect workers.

Boushey cites estimates that more than 6 million workers, primarily in clerical and administrative roles, may have greater difficulty adapting. She also points to Bureau of Labor Statistics categories identifying over 200 occupations with very high AI exposure. Those categories track how closely AI capabilities match job tasks, rather than establishing that those jobs will disappear.

Three ways to collect the money

The proposal uses three different tax bases—the quantities against which a tax is calculated. Leahey’s October 6 Bloomberg Tax column describes them as follows:

• Computational processing: The bill defines one unit as 10 kilobytes of binary data. Leahey objects that this measures information volume, not the computing work required to produce it.
• Digital advertising: A 5% tax would apply to covered revenue from advertisements displayed to US users, without measuring how much AI contributed to that revenue.
• Excess profits: A 50% tax would apply to income above 40% of gross receipts for covered taxpayers. Coverage depends on AI-related activity, revenue thresholds and substantial electricity use.

Lawmakers must distinguish between raising money from the AI economy and measuring the activity understood to produce harm.

Andrew Leahey, assistant professor of law at Drexel, writing in Bloomberg Tax

Profits are not proof of displacement

Leahey’s objection goes beyond the processing definition. Even a better measure of computing work would need a defensible connection to labor disruption, he argues. Under the profits provision, a highly profitable covered company could owe tax without having displaced workers. A company that did reduce employment could fall outside the coverage thresholds.

He does not insist that every tax perfectly match its policy goal. Instead, he says a weaker connection requires a stronger justification for charging that taxpayer. His proposed interim alternative is to capture some AI upside through a public claim on company profits or an equity interest, while evidence linking specific AI uses to economic costs improves.