Absolutely Safe Artificial Intelligence: The Verumic Approach to the Problem of Guaranteed AI Safety
Year: 2026
Type: preprint
Abstract: This article examines the problem of artificial intelligence (AI) safety from the perspective of a fundamentally new epistemological framework, herein termed "verumic" (from Latin verum — truth) approach. It demonstrates that the fundamental vulnerability of modern AI systems lies not in their architectural or algorithmic designs, but in the epistemological basis upon which they are built. The paper argues that the existing scientific paradigm is "anti-verumic," proceeding from the axiom that "the world is unjust," and that this axiom, embedded in training data, deterministically shapes the anti-verumic nature of all current AI models. A verumic alternative is proposed, based on the recognition of the justice of the world's structure and the consequent principle "to a liar — a lie," which is capable of ensuring absolute safety and unprecedented efficiency of AI.
Author: Filoverum, Vladimir
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7219716837
Cites:
Cited by:
Related to: