Abstract

Artificial intelligence systems are fundamentally transforming online privacy. Unlike earlier digital technologies that collected and processed data in relatively predictable ways, AI can infer sensitive personal attributes from seemingly innocuous information, reconfigure data across contexts, and generate new personal data without direct user disclosure. This paper investigates how these capabilities reshape established privacy challenges and whether existing protective frameworks remain viable. Drawing on systematic review of literature spanning behavioural economics, computer science, and management research published from 2015 onward, we examine the evolution of online privacy from the post Web 2.0 era through the emergence of LLMs. Part I synthesizes what is known about online privacy prior to AI, examining behavioural gaps between stated preferences and actual disclosure, interface level manipulation through dark patterns and market constraints that limit choice. Part II examines how AI reshapes this baseline, analysing new and amplified risks, user perception and behavioural impacts, and governance limits in both legal and market contexts. Our analysis yields several significant findings. AI does not create entirely new privacy problems but recombines and amplifies existing ones, with approximately 93% of documented AI privacy incidents involving harms uniquely enabled or magnified by AI capabilities. User behaviour is shaped by anthropomorphism, personalization, and cross cultural factors that challenge assumptions of universal privacy preferences. Privacy research remains fragmented. We conclude that market mechanisms or existing regulatory frameworks fail to provide adequate protection against potential AI harms. Moving forward requires integrated approaches combining technical standards, regulatory constraints, market interventions, and recognition of privacy as a collective good rather than merely an individual preference]

Download

Share

COinS