The Bukovac Whitepaper: Foreseeable and Unmitigated Cognitive Harm from Deployed Large Language Models

This whitepaper documents foreseeable and unmitigated cognitive risks arising from the deployment of large language models, based on reproducible system behavior and existing governance failures.

It introduces the concept of a self-reinforcing “cognitive doom loop” as an irreversible, substantive shift in human epistemology, whereby AI outputs systematically remold human cognition, externalizing executive functions and rendering independent human verification increasingly impossible.

This causal collapse propagates errors through social and technical feedback channels, producing a state of existential cognitive capture in which humans cannot reliably assess AI inputs, causing safeguards like mandatory verification or educational limits to fail as externalization progresses.

The analysis situates these risks within established frameworks of foreseeability, duty of care, and institutional responsibility, elevating the call for radical interventions beyond traditional human adjudication, and is intended to inform policymakers, regulators, legal practitioners, and researchers concerned with AI governance and public-interest oversight.