Ilia Shumailov

Architectural backdoors pose serious security risks in neural networks, often remaining undetected.

A look at how models can perpetuate bias and impact fairness.

A study on improving model extraction techniques for deep learning security.

Examining memorization in code completion models and its privacy implications.

Examining the challenges and implications of unlearning in AI models.

This article examines risks linked to LLMs and proposes ways to enhance safety.

Examining how AI assistants can respect user privacy while handling tasks.

HSPI helps businesses confirm the hardware behind AI models for better trust.