Containment, Not Catastrophe

Artificial intelligence is increasingly shaping the fundamental pillars of society, from the allocation of capital in finance and the dissemination of knowledge in education, to the provision of care in health and the exercise of power in politics, the critical question we face is no longer if AI should be regulated, but how. The imperative for regulation isn't born from speculative sci-fi nightmares of rogue sentient machines threatening humanity. Rather, it is driven by concrete, measurable harms that are already manifesting in our daily lives.

The scale and speed of AI deployment across society has outpaced our regulatory frameworks by decades. Consider that the same algorithms deciding mortgage approvals, criminal sentencing recommendations, and medical diagnoses operate with less oversight than the approval process for a new breakfast cereal.

Documented Harms So Far

Bias in Critical Systems

Recent studies have documented how AI systems systematically disadvantage marginalized communities. The landmark Gender Shades study out of MIT found commercial facial-analysis systems misclassified darker-skinned women at error rates as high as 34.7%, compared to error rates near zero for lighter-skinned men. The study's influence pushed major vendors to measurably improve accuracy for underserved groups, proof that scrutiny works, but also proof the harm was real and already shipping in commercial products.

Hiring algorithms trained on historical data raise similar concerns. In Mobley v. Workday, a federal court in 2026 allowed discrimination claims to proceed against an AI-powered applicant-screening tool, with plaintiffs alleging the system disadvantaged candidates over 40 and, in earlier filings, on the basis of race and disability. It's one of several active cases, alongside a widely reported ACLU complaint over an AI interview tool used against a deaf, Indigenous applicant, testing whether existing discrimination law can keep pace with automated hiring decisions made at scale.

The Surveillance Economy

Modern AI surveillance extends far beyond traditional security cameras. Educational institutions deploy emotion recognition software to monitor student engagement, creating psychological profiles that follow students throughout their academic careers. Workplace surveillance systems track employee productivity, bathroom breaks, and social interactions, often with far less transparency than the hiring algorithms above, because monitoring tools are rarely subject to the same legal scrutiny as decisions like termination or promotion.

Regulation in 2026

Talk of AI regulation isn't hypothetical anymore, it's actively being written, delayed, and litigated in real time, and the picture is more mixed than either AI optimists or AI alarmists tend to admit.

In the European Union, most remaining provisions of the EU AI Act took effect on August 2, 2026, including transparency duties and enforcement powers over general-purpose AI providers, with maximum fines reaching €35 million or 7% of global turnover. But under the Commission's "Digital Omnibus" initiative, obligations for high-risk AI systems were deferred to December 2027 and August 2028, a reminder that even the world's most comprehensive AI law is still being negotiated in practice, not just on paper.

In the United States, there is no federal AI law of comparable scope. Instead, regulation is happening state by state, and fast: as of early 2026, lawmakers in 45 states had introduced over 1,500 AI-related bills, more than in all of 2024 combined, according to the multistate.ai legislation tracker. The approaches vary sharply. California's SB 942 requires generative AI providers to offer watermarking and content-detection tools. Colorado replaced its original AI Act with a narrower automated-decision-making statute effective 2027, requiring consumer notice and a right to human review for high-stakes decisions. Texas, by contrast, passed a Responsible AI Governance Act that mostly limits obligations to government use of AI, leaving the private sector largely untouched. That patchwork is itself a preview of the coordination problem discussed below.

9 Ways to Regulate AI

1. Audits and Transparency

Mandate regular algorithmic audits for AI systems used in high-stakes decisions. These audits should examine training data, decision-making processes, and outcomes for bias and fairness, the kind of independent scrutiny that identified the accuracy disparities in the Gender Shades study years before regulation caught up.

2. Data Rights and Digital Dignity

Implement comprehensive data rights legislation that includes the right to algorithmic explanation, the right to human review of automated decisions, and the right to data portability. Colorado's revised automated-decision-making law is an early, narrow example of this principle in statute.

3. Algorithmic Labor Rights

Establish labor rights for AI workers, including fair wages, safe working conditions, and protection from psychological harm, a category that includes both the human raters and moderators training AI systems and the workers whose jobs are being restructured around them.

4. Public Interest AI Development

Establish public AI research institutions and require public benefit considerations in AI development, rather than leaving the direction of the technology entirely to a handful of commercial labs.

5. Sectoral AI Governance

Develop sector-specific AI governance frameworks that address the particular risks and requirements of each domain, healthcare, hiring, and criminal justice each need different guardrails, which is why Pennsylvania is pursuing healthcare-specific AI rules while New Jersey focuses on employment.

6. Public Say in AI Rules

Create participatory governance mechanisms that give communities a voice in AI decisions that affect them, rather than leaving governance entirely to legislators and vendors negotiating without public input.

7. Liability and Redress

Establish clear liability frameworks for AI harms, including strict liability for certain high-risk AI applications. Active litigation like Mobley v. Workday is, in effect, courts trying to build this framework case by case in the absence of clear statutes.

8. Global Cooperation

Develop international AI governance standards and cooperation mechanisms. The divergence between the EU's comprehensive-but-delayed approach and the US's fragmented state-by-state approach shows what happens without them: companies operating across borders face genuinely different rules depending on jurisdiction.

9. Adaptive Regulation

Create adaptive regulatory frameworks that can evolve with technology, the EU's own decision to defer high-risk provisions by more than a year shows that even well-resourced regulators are still calibrating timelines against real-world implementation capacity.

What It Means for Business

For companies building AI-powered products rather than legislating them, the practical implication of this patchwork is that compliance can no longer be an afterthought bolted on before a launch. A product using AI in hiring, credit, healthcare, or education needs to track requirements across every jurisdiction it operates in, not just its headquarters, and the safest working assumption in 2026 is that the requirements will keep changing under you. Building in audit logging, explainability, and human-review pathways from the start is significantly cheaper than retrofitting them once a regulator or a plaintiff's attorney asks for them.

What It Means for You

Most people encountering AI-driven decisions, a loan application, a job screening, an insurance quote, still have very little visibility into when AI was involved at all, let alone how it reached its conclusion. The regulatory patchwork described above is starting to change that, but unevenly: a resident of Colorado will have meaningfully different rights to an explanation and human review than a resident of Texas, simply based on where they live, at least until federal legislation or interstate coordination catches up. Until that gap closes, the most practical protection available to most people is simply asking, directly, whether a decision that affects them involved an automated system, a question that a growing number of state laws now give people an explicit right to ask.

Frequently Asked Questions

Is there a federal AI law in the United States?

Not yet, as of 2026. Regulation is happening primarily at the state level, which is why requirements vary significantly depending on where a company operates and which state's residents it serves.

What happens if a company doesn't comply with the EU AI Act?

Penalties can reach €35 million or 7% of global annual turnover, whichever is higher, a maximum fine that exceeds even the GDPR's, reflecting how seriously EU regulators are treating high-risk AI use cases.

Does regulation slow innovation?

It's a real trade-off, but not an automatic one. Well-designed sectoral rules, the kind we discuss above, tend to concentrate compliance burden on genuinely high-risk uses (hiring, healthcare, credit) rather than blanket restrictions, which is closer to the direction most 2026 state laws are actually taking.

How can a business track which AI regulations apply to it?

Given how fast the state-by-state landscape is moving in the US alongside the EU's own shifting timeline, most companies operating across multiple jurisdictions rely on a dedicated legal or compliance function rather than tracking changes ad hoc. For smaller teams without that resource, following an established tracker like the ones cited above and reviewing it quarterly is a reasonable minimum, especially for any product touching hiring, credit, healthcare, or education.

Conclusion

We stand at a critical juncture in human history. The choices we make about AI governance in the coming years will shape the trajectory of technological development and social progress for generations to come. As we've written about separately, public trust in AI remains deeply divided, and that divide is itself part of what regulators are responding to. The nine strategies outlined here, several of which are already showing up in early, imperfect form in real statutes and court cases, provide a roadmap for containing AI's risks while preserving its benefits.