Adversarial Fragility and Language Vulnerability in Clinical AI: A Systematic Audit of Diagnostic Collapse Under Imperceptible Perturbations and Cross-Lingual Drift in Low-Resource Healthcare Settings

TL;DR

This paper systematically audits clinical AI systems revealing their vulnerability to imperceptible adversarial image perturbations and cross-lingual diagnostic drift, especially in low-resource healthcare settings.

Contribution

It introduces the first dual audit of adversarial fragility and language vulnerability in clinical AI, demonstrating significant accuracy drops under realistic perturbations and linguistic variations.

Findings

Diagnostic accuracy drops from 89.3% to 62.0% under imperceptible image perturbations.

Language-based accuracy decreases significantly across Pidgin and Yoruba-inflected English.

Standard defenses like Gaussian smoothing and ensemble voting do not restore safety.

Abstract

Current clinical artificial intelligence (AI) systems are evaluated almost exclusively on clean, standardised, English-language inputs, conditions that do not reflect the realities of healthcare delivery in low-resource settings. This study presents the first systematic dual audit of two orthogonal safety vulnerabilities in clinical AI: adversarial image fragility and cross-lingual diagnostic drift. Using DenseNet121, the architecture underlying CheXNet, fine-tuned on the COVID-QU-Ex chest X-ray dataset (85,318 images; COVID-19, Non-COVID Pneumonia, Normal), we demonstrate that diagnostic accuracy collapses from 89.3% to 62.0% under a Fast Gradient Method (FGM) perturbation of epsilon=0.021, a magnitude imperceptible to the human eye. Standard defensive strategies including Gaussian smoothing and ensemble voting failed to restore clinical safety. In a parallel language fragility experiment, we tested Llama3.1:8b and NatLAS (N-ATLAS) on 20 COVID-19 clinical cases presented in Standard English, Nigerian Pidgin (Naija), and Yoruba-inflected English. Both models exhibited significant accuracy degradation: Llama3.1:8b dropped from 80.0% to 65.0% on Pidgin; NatLAS, an African-context model, collapsed from 85.0% to 55.0%, with diagnosis consistency falling to 50%. These findings establish a quantitative failure envelope for clinical AI under conditions representative of Primary Health Centre (PHC) deployment in Nigeria, and motivate urgent calls for adversarially hardened, linguistically inclusive clinical AI architectures.