Michael Backes

This article discusses the privacy concerns of using GPT models in cloud settings.

Examination of jailbreak attacks shows weaknesses in language model safety.

Custom LLMs raise safety concerns, particularly with instruction backdoor attacks.

Diverse samples enhance the effectiveness of machine learning model theft.

A new framework assesses the effectiveness of image safety classifiers against harmful content.

Inductive GNNs face privacy threats from link stealing attacks.

A new framework controls in-context learning to prevent misuse in AI models.

Examining the threats posed by autonomous language model agents and their weaknesses.

Examining the effects of updates on safety, bias, and authenticity in image generation.

Examining how important data points attract more security risks in machine learning.

Examining how SSL models memorize data points and its implications.