Towards Assuring EU AI Act Compliance and Adversarial Robustness of LLMs

TL;DR

This paper presents a framework combining ontologies, assurance cases, and factsheets to help ensure large language models comply with EU AI regulations and are resilient against adversarial attacks.

Contribution

It introduces a novel framework that aids engineers and stakeholders in documenting and verifying LLM compliance and security, addressing current implementation challenges.

Findings

Framework supports understanding and documenting AI compliance.

Enhances adversarial robustness of LLMs.

Facilitates regulatory adherence and security assurance.

Abstract

Large language models are prone to misuse and vulnerable to security threats, raising significant safety and security concerns. The European Union's Artificial Intelligence Act seeks to enforce AI robustness in certain contexts, but faces implementation challenges due to the lack of standards, complexity of LLMs and emerging security vulnerabilities. Our research introduces a framework using ontologies, assurance cases, and factsheets to support engineers and stakeholders in understanding and documenting AI system compliance and security regarding adversarial robustness. This approach aims to ensure that LLMs adhere to regulatory standards and are equipped to counter potential threats.