The panoptic sort, originally theorized by Oscar Gandy in 1993, is the mechanism by which individuals are classified into categories based on data about their behavior, demographics, and characteristics—and the classification determines the opportunities, prices, services, and treatment they receive. The sort is panoptic in Foucault's sense: it operates through asymmetric visibility, where the sorter sees the sorted but the sorted cannot see the sorter, cannot know the criteria, cannot contest the classification. Zuboff incorporated Gandy's framework into her analysis of surveillance capitalism, showing how behavioral surplus is processed into sorting mechanisms that operate invisibly and at scale. The AI moment introduces a new dimension: cognitive sorting, in which the classification criterion is not demographic profile or purchase history but the individual's relationship to AI itself—how effectively they use the tools, how well they evaluate output, how skillfully they develop the evaluative intellective skill that AI-augmented work demands. The sorting produces consequences through differential opportunity: AI-fluent workers receive interesting projects, visibility, advancement; AI-unable or AI-resistant workers receive maintenance tasks, legacy support, marginalization. The mechanism is sufficiently diffuse that the sorted individual may not recognize the sorting as it happens.
The contemporary panoptic sort operates on multiple levels simultaneously. At the individual level within organizations, workers are sorted by their capacity to use AI tools effectively—a sorting that may not be explicit (formal performance categories) but is visible in the allocation of opportunity, responsibility, and advancement. At the organizational level, firms are sorted by their adaptation to AI—the software death cross is a market-scale sorting event, with a trillion dollars redistributed according to which companies possessed ecosystem value that AI could not replicate. At the national level, the sorting operates through the distribution of AI capability, infrastructure, and institutional design—nations that invest in AI-capable workforces and governance frameworks are sorted into the upper tier of the global knowledge economy; nations that do not are sorted out.
The cognitive sort is more intimate and more consequential than demographic sorting because it classifies people by capability—by how they think, how they solve problems, how they exercise judgment under uncertainty. If platforms that provide AI tools collect and process data on user interaction patterns—which prompts produce good results, which outputs are accepted or rejected, how long users spend evaluating, what characterizes their successful problem-solving—then the platforms possess knowledge about user capability that could be monetized in hiring decisions, performance evaluations, lending assessments, insurance underwriting. The cognitive sort would be a panoptic sort operating on thinking itself, and its invisibility would be its power: the sorted individual would experience differential outcomes—the job offer, the declined loan, the higher premium—without understanding that the outcome was produced by a classification system whose inputs were the individual's interactions with AI tools.
Zuboff's framework suggests that the governance challenge is making the sort visible. As long as the sorting operates beneath awareness, it cannot be contested. Workers experience differential opportunity without recognizing that they are being systematically classified. Organizations experience competitive disadvantage without understanding that they are being sorted by criteria they did not choose. The visibility is the prerequisite for contestability, and contestability is the prerequisite for democratic governance of the mechanism. Without transparency requirements—disclosure of when AI interaction data is used in consequential decisions, explainability of the criteria by which classification operates—the cognitive sort will reproduce the pattern of every previous panoptic sort: invisible extraction producing unaccountable consequences borne by people who cannot challenge the mechanism that produced them.
Oscar Gandy introduced the panoptic sort in The Panoptic Sort: A Political Economy of Personal Information (1993), extending Foucault's panopticon analysis to information technology and data-driven classification. Zuboff adopted and extended the concept across her surveillance capitalism work, showing how platforms had industrialized sorting at scales Gandy had not anticipated. The AI-era extension—cognitive sorting by interaction pattern—is this volume's contribution, reading the Zuboff and Gandy frameworks forward into the moment when the data being sorted is not demographic or behavioral but cognitive.
Classification by Data-Derived Categories. The sort operates not on self-reported identity but on patterns extracted from behavior—and in AI, from cognition.
Asymmetric Visibility. The sorter sees; the sorted cannot—opacity is structural, not incidental, because visibility would enable contestation.
Consequences Without Consent. The sorted individual experiences outcomes (differential opportunity, pricing, access) without consenting to the classification that produced them.
Cognitive Sorting in AI. The new frontier: classification by thinking capability revealed through AI interactions—more intimate and consequential than any demographic category.
Governance Requires Visibility. The sort cannot be democratically governed until it is made visible—transparency and explainability are prerequisites for contestability.