RSS

Dmitrii Savchenko

Karma: 0

Independent researcher working on architectural limits of verifiability

in adversarial AI settings.

Core question: can an environment be designed where an autonomous agent

structurally cannot distinguish a true result from a plausible one?

I believe yes, and have a formal construction + conceptual validation

against 10 frontier models. This is intermediate work — single-query

case is proven, adaptive multi-query remains open.

Background: 25 years in business, with AI-focused work since 2018

(industrial automation, conversational AI, enterprise management).

Transitioning into foundational safety research. No academic affiliation.

Preprint: https://​​doi.org/​​10.5281/​​zenodo.21261173

Sandbox for replication: https://​​hybra.ru/​​mirage/​​

No entries.