ABSTRACT
Privacy law was built to protect data. But the emergence of persistent, context-aggregating agentic AI systems poses a threat that data protection cannot address: the reduction of a person’s plural, context-dependent identity to a single statistical composite. This essay introduces the concept of cognitive identity capture-the process by which an agentic AI system, through the accumulation of personal context across all domains of a person’s life, constructs and imposes a unified model of who that person is, undermining their capacity to be different people in different contexts. Drawing on Kant’s account of autonomy, Clark and Chalmers’ extended mind thesis, Nissenbaum’s contextual integrity framework, Illich’s distinction between convivial and non-convivial tools, and Damasio and McGaugh’s neuroscience of selective memory, the essay argues that cognitive identity capture constitutes a distinct harm – one that is structural, architectural, and prior to any question of data ownership or compensation. Against Lanier’s Data Dignity proposal, which addresses the terms of extraction, this essay addresses the need for extraction itself. It proposes three rights that current legal frameworks do not provide: the right to be plural, the right to be unlearned, and the right to determine, in each context, which version of oneself is present. It concludes with a set of architectural conditions-local-first context storage, identity-bounded domains, model-agnostic context management, and explicit bridges-that define what a convivial AI system requires. The essay is written in the first person, as a position piece in the Open Forum tradition, grounded in both academic literature and the author’s practice as an AI strategist, educator, and researcher.
Veldsink, Jan, Privacy Is Not Enough: On Identity, Agentic AI, and What We Are Actually Losing (August 5, 2026).
Leave a Reply