Response to Request for Information on Regulatory Reform on Artificial Intelligence

Notice of Request for Information; Regulatory Reform on Artificial Intelligence
Agency: Office of Science and Technology Policy
Comment Period Opens: September 26, 2025
Comment Period Closes: October 27, 2025
Comment Submitted: October 27, 2025
Docket No. OSTP-TECH-2025-0067

This letter responds to the OSTP’s request for feedback on regulatory reforms to support responsible artificial intelligence. I am a research fellow at the Mercatus Center at George Mason University. My research examines how the federal tax code shapes firms’ choices between machines and people, and how policy can accelerate worker upskilling during rapid AI diffusion. Drawing on my recent policy brief, I present six concrete reforms.^[1] For each, I offer a rationale and a concise recommendation.

Why Reform Is Needed

• Federal tax rules front-load deductions for equipment (e.g., Internal Revenue Code § 168(k) bonus depreciation) while rationing or disallowing many bona fide training expenses under §§ 162 and 127.

• This aspect of tax law creates a policy bias toward physical capital over human capital, depressing firm-led upskilling precisely when AI complements are most valuable.

• The bias is driven by six legacy restrictions: the “new trade or business” and “minimum educational requirements” tests under § 162; the 5 percent owner cap, nondiscrimination testing, and a nominal $5,250 cap under § 127; and the lack of full expensing for training.

• Neutral, pro-worker AI adoption requires treating investments in people on par with investments in machines.

Reform 1—Modernize § 162: “New Trade or Business”

Key recommendation: Allow deductions for job-related education and training that develop skills for evolving work within the firm’s line of business, including training needed to use or supervise AI systems.

Why it matters: The “new trade or business” bar, developed for a slower economy, now misclassifies natural, AI-driven task shifts as a move into a different trade, penalizing foundational upskilling and discouraging investment in worker complements to AI.

Reform 2—Modernize § 162: “Minimum Educational Requirements”

Key recommendation: Allow deductions for training that establishes baseline competence for AI-mediated tasks (e.g., safe operation, data quality, oversight) when connected to the firm’s existing business activities.

Why it matters: Disallowing training that meets a “minimum requirement” blocks entry-level AI capabilities that make hardware and software investments safe and productive, especially for smaller firms.

Reform 3—Update § 127: Remove the 5% Owner Limitation

Key recommendation: Eliminate the 5% owner cap so owners of small and family firms can participate in qualified educational assistance programs (QEAPs).

Why it matters: The cap excludes the dominant US business forms—sole proprietorships and family LLCs—which are precisely the types of businesses where entrepreneurs must retrain themselves and their teams to adopt AI tools.

Reform 4—Update § 127: Replace Rigid Nondiscrimination Ratios with Anti-abuse Standards

Key recommendation: Replace mechanical ratio tests of high-income vs. low-income worker take-up with anti-abuse enforcement that preserves broad availability of AI education and training opportunities.

Why it matters: When take-up of these training programs differs across pay bands because of awareness or scheduling constraints, ratio tests encourage employers to discontinue programs, leaving workers unserved and slowing AI-related skill formation.

Reform 5—Update § 127: Lift the $5,250 Cap

Key recommendation: Lift the $5,250 annual cap on educational assistance.

Why it matters: Set in 1986, the cap has lost much of its real value because of inflation. Many credible AI courses, certifications, and supervised training programs exceed that amount, making the cap a binding constraint on meaningful skill development. 

Reform 6—Human-Capital Parity: Full and Immediate Expensing for Training

Key recommendation: Extend full and immediate expensing to bona fide job-related training, achieving parity with § 168(k) equipment expensing.

Why it matters: If an AI server is expensed in year one while complementary training is rationed, policy tilts toward hardware without the human capabilities that generate productivity and safety. Parity improves neutrality, speeds adoption with fewer displacements, and—like bonus depreciation—mainly shifts deduction timing.

Conclusion

Responsible AI requires responsible incentives. Modernizing §§ 162 and 127 and granting full expensing to job-related training would restore neutrality between machines and people, reduce displacement risk, and accelerate safe, productivity-enhancing AI adoption—especially among small businesses. These are low-cost, high-return reforms grounded in established tax principles and evidence on firm-led training.