Article

From Intent to Impact: Equal Protection in the Age of Artificial Intelligence

AI is pervading governmental decision-making across a wide range of sectors and throughout all levels of government. Yet the hallmarks of modern equal protection doctrine—discriminatory intent and formal classification—are increasingly inapplicable to AI-inflected decision-making. The opacity, complexity, and autonomy of AI models make it nearly impossible to attribute discriminatory intent to those who build or rely on them; meanwhile, such models can be designed to operate free of the formal classifications that the doctrine prohibits. Evolving interpretations of the Equal Protection Clause have thus rendered AI largely exempt from constitutional scrutiny, even as AI may produce disparities in group outcomes by relying on data correlated with race and other protected characteristics.

What should one make of this growing disconnect? One response is to celebrate AI’s “colorblindness” as a triumph of the anti-classification vision. While we understand this response, we think it is misguided. The historical understanding of equal protection has long embodied an anti-subordination sensibility as well. If the doctrine is to remain a meaningful constraint on government in the age of AI, it should shift toward an impact-oriented framework more attuned to systemic inequality. Admittedly, the Court may simply decide that algorithmic governance effectively lies beyond the reach of the Equal Protection Clause. But the costs of such a decision will only grow as AI continues to pervade governmental decisionmaking.

* Ralph Richard Banks is the Jackson Eli Reynolds Professor of Law at Stanford Law School.

* Victor Y. Wu is a Ph.D. Candidate in Political Science at Stanford University and a Data Science Scholar at the Stanford Institute for Human-Centered AI. He received his J.D. from Stanford Law School in 2025.
The authors are grateful for helpful comments and feedback from Bradley Joondeph, Pauline Kim, Kathy Mack, Julian Nyarko, Nate Persily, and participants at the 2026 Law and Society Association Annual Meeting.

The full text of this Article is available to download as a PDF.