Work

LLM Security Re-framed: An Entropy-Led, Physics-Grounded Critique of the "Vulnerability" Narrative

Year: 2025

Type: article

Abstract: LLMs don’t have bugs—they have exhaust.We pipe the steam, measure the joules, and prove the ‘vulnerabilities’ are just thermodynamic inevitabilities.Large Language Models (LLMs) are not broken they are working exactly as designed: autoregressive, overparameterized, statistical heat engines converting data entropy into token sequences. Their so-called “vulnerabilities” — hallucinations, data leakage, prompt injection — are not implementation flaws, but mathematical and physical inevitabilities arising from their core architecture.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.17173812

Author: Sunny, Thakur

Language: English

Topic: Adversarial Robustness in Machine Learning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7082636707

FWCI:

Cites:

Cited by:

Related to: