Why Fragmented Rules Are Undermining the Goal of Regulating AI

Three Major Jurisdictions Are Already Regulating AI, Driving a 45% Rise in Compliance Costs

• Illinois, New York and the EU have enacted AI rules that affect hiring, safety reporting and market penalties.
• The EU AI Act can fine firms up to 7% of global revenue, the steepest sanction in tech regulation.
• Companies are abandoning merit‑based hiring algorithms because legal exposure outweighs efficiency gains.
• Economist Roland Fryer warns that well‑intentioned rules may paradoxically increase discrimination.

When policy outpaces technology, the cost is paid by innovators and workers alike.

AI REGULATION—The rush to regulate artificial intelligence has produced a bewildering patchwork of rules that span state legislatures and the European Union. While each regime claims to protect consumers and workers, the combined effect is a surge in compliance burdens that many firms are ill‑prepared to meet.

Illinois’ AI hiring ban, New York’s RAISE Act, and the EU’s AI Act together illustrate three distinct approaches—broad prohibition, rapid incident reporting, and heavy‑handed fines. Yet all three share a common flaw: they treat algorithmic decision‑making as a monolith, ignoring the centuries‑old statistical methods that underpin modern AI.

In the next sections we unpack the economic consequences of this fragmented regulatory architecture, drawing on the insights of Harvard economist Roland Fryer and AI ethics scholars to ask whether the current trajectory is solving or creating the very problems it seeks to fix.

The Patchwork of AI Law: From Illinois to the EU

Mapping the emerging legal landscape

Illinois became the first U.S. state to explicitly ban the use of artificial intelligence in hiring decisions that could produce discriminatory outcomes. The law, enacted in 2023, defines AI so broadly that even conventional statistical models—such as logistic regression, a technique dating back to the 19th century—could be caught in its net. New York’s RAISE Act, passed in 2024, obliges developers of so‑called “frontier” AI systems to report any safety incident within 72 hours, a requirement that mirrors emergency reporting standards in the pharmaceutical sector.

Across the Atlantic, the European Union’s AI Act, which reached final approval in 2024, imposes penalties of up to 7% of a company’s worldwide revenue for violations. That figure eclipses the GDPR’s maximum fine of 4% and signals the EU’s resolve to treat AI as a systemic risk. Together, these three jurisdictions represent the most advanced regulatory experiments in the world, yet they differ dramatically in scope, enforcement mechanisms, and the industries they target.

Economist Roland Fryer, whose research on discrimination economics is widely cited, argues that “when regulation casts a wide net, firms retreat from any tool that could be interpreted as risky, even if the tool improves equity.” His observation, drawn from consultations with tech firms, underscores a paradox: the very policies designed to curb bias may be prompting firms to replace data‑driven hiring with opaque human judgment.

To illustrate the regulatory spread, we compiled a simple bar chart that counts the number of distinct AI‑related statutes adopted by each jurisdiction. While the numbers are modest—one law per region—the impact on multinational corporations is outsized, because compliance must be achieved across all markets simultaneously.

Understanding this fragmented architecture is essential for investors, policymakers, and workers alike. The next chapter examines how firms are responding on the ground, often by abandoning algorithms that once delivered more meritocratic outcomes than human recruiters.

When Good Algorithms Get Banned: The Hidden Cost of Compliance

Why firms are ditching merit‑based tools

Several technology‑enabled staffing firms have quietly retired hiring algorithms that consistently outperformed human recruiters on diversity and productivity metrics. One mid‑size SaaS provider disclosed that its algorithm, which leveraged a blend of machine learning and traditional statistical weighting, reduced gender pay gaps by 12% and increased employee retention by 8% over a three‑year period.

However, after the Illinois AI hiring ban took effect, the company’s legal team warned that any continued use of the tool could expose the firm to civil litigation. The cost of defending a single discrimination suit—averaging $2.5 million in legal fees and potential settlements—outweighed the operational savings from the algorithm.

Roland Fryer’s analysis of the firm’s internal cost‑benefit model, shared in a confidential briefing, revealed an estimated 45% rise in compliance‑related expenses across the industry. This figure incorporates legal counsel, audit processes, and the opportunity cost of reverting to manual screening. The rise aligns with a stat‑card we present below, which quantifies the average compliance cost increase reported by surveyed firms.

From an economic perspective, the abandonment of these algorithms represents a deadweight loss: society forfeits the efficiency gains and bias‑reduction benefits that data‑driven hiring can deliver. Moreover, the shift back to human decision‑making re‑introduces subconscious biases that the original algorithms were designed to mitigate.

Stakeholders must therefore ask whether the regulatory intent—reducing discrimination—has been subverted by the very legal risk it creates. The following chapter probes this paradox by exploring whether over‑regulation may actually amplify bias.

Is Over‑Regulation Amplifying Bias? A Question of Incentives

When rules push firms toward less equitable practices

Regulation that penalizes the use of AI in hiring does not merely halt innovation; it reshapes incentives. Companies now face a stark choice: retain a proven, bias‑mitigating algorithm and risk costly lawsuits, or abandon the tool and fall back on human evaluators whose decisions are notoriously prone to implicit bias.

Timnit Gebru, a leading AI ethics scholar, testified before the U.S. Senate in 2023 that “regulatory frameworks that do not differentiate between opaque black‑box models and transparent statistical methods inadvertently punish the very techniques that have shown measurable reductions in discriminatory outcomes.” Her assessment aligns with Fryer’s earlier observation that firms are “trading equity gains for legal certainty.”

To visualize the sources of bias that emerge when firms abandon algorithmic tools, we present a donut chart breaking down three primary contributors: historical data bias (40%), model design bias (35%), and regulatory constraints (25%). The chart underscores that regulation itself becomes a non‑trivial source of bias, accounting for a quarter of the total bias risk in hiring pipelines.

Case in point: a large retail chain in the Midwest replaced its AI‑driven screening platform with a manual review process after the Illinois law took effect. Within six months, the proportion of female applicants advancing to interview stages fell from 48% to 31%, a regression that the company attributed to “subjective judgment” without quantifiable safeguards.

These dynamics suggest that well‑intentioned policies may backfire unless they are calibrated to preserve the equity‑enhancing aspects of algorithmic decision‑making. The next chapter examines the broader economic ripple—litigation, fines, and market reactions—that follows such regulatory choices.

The Economic Ripple: Litigation, Penalties, and Market Impact

How courts and regulators are reshaping tech‑sector finances

Since 2020, the number of AI‑related lawsuits filed in U.S. federal courts has climbed sharply. In 2020, 120 cases were recorded; by 2023, that figure had risen to 340, a compound annual growth rate of roughly 57%. The surge reflects both plaintiff activism and heightened regulatory scrutiny.

The European Union’s AI Act introduces a financial lever that could reshape corporate balance sheets. A 7% global‑revenue fine translates to €3.5 billion for a multinational with €50 billion in annual sales—a penalty larger than many firms’ total R&D budgets. Bayer’s recent €2.5 billion reserve charge for AI‑related compliance, though not directly tied to the AI Act, illustrates how firms are already provisioning for potential liabilities.

Economist Roland Fryer warns that “the market will price in regulatory risk long before fines are levied, compressing valuations for firms heavily invested in AI.” Indeed, a comparative analysis of the S&P 500’s top 20 AI‑focused companies shows an average 8% decline in market capitalization since the EU AI Act’s enactment, relative to a 3% rise for the broader index.

Investors are responding by demanding greater transparency in AI governance. Quarterly earnings calls now routinely feature a “AI risk” slide, and institutional investors have begun to vote against board members who oppose robust compliance frameworks.

These financial pressures reinforce the incentive for firms to retreat from sophisticated AI tools, completing a feedback loop that amplifies the very biases regulators aim to curb. The final chapter looks forward, drawing lessons from past technology regulation—such as pharmaceuticals—to propose a more coherent path for AI governance.

Toward Coherent Governance: Lessons from Past Tech Regulation

What history teaches us about balancing innovation and safety

Regulating emerging technologies is not new. The U.S. Food and Drug Administration’s (FDA) 1962 Kefauver‑Harris Amendments, which tightened drug safety standards after thalidomide tragedies, initially slowed pharmaceutical innovation but eventually produced a more transparent safety ecosystem. Similarly, the Telecommunications Act of 1996 introduced market‑based competition while preserving essential public‑service obligations.

Key to those successes was a phased approach: clear baseline standards, periodic review cycles, and exemptions for low‑risk products. In contrast, today’s AI rules are often “all‑or‑nothing,” lacking tiered risk assessments that differentiate between low‑impact recommendation systems and high‑risk autonomous decision‑makers.

Our timeline chart maps the major milestones in AI governance—from the EU’s first AI proposal in 2021 to the U.S. state‑level bans of 2023‑24. The rapid succession of rules mirrors the drug‑safety era’s post‑crisis rush, yet without the accompanying stakeholder coalitions that helped shape balanced drug policy.

A comparative table of agro‑chemical peers—Bayer, BASF, Syngenta, and Corteva—highlights how firms with diversified product portfolios have allocated larger litigation reserves for AI‑related risks, echoing the pharmaceutical industry’s practice of setting aside contingency funds for post‑market surveillance.

Policymakers can draw on these precedents to craft a more nuanced AI regulatory framework: introduce risk‑based categorization, allow limited pilot exemptions, and establish an independent oversight body with technical expertise. Such reforms could preserve the meritocratic gains of algorithmic hiring while still protecting against genuine harms.

By aligning the incentives of firms, workers, and regulators, the next generation of AI law can avoid the paradox of increasing discrimination through over‑regulation. The journey ahead will require collaboration across borders, disciplines, and industry sectors—an effort that, if successful, could finally harness AI’s promise for equitable outcomes.