AI Act and Large Language Models (LLMs): When critical issues and privacy impact require human and ethical oversight

TL;DR

The paper analyzes the EU AI Act with a focus on human and ethical oversight and privacy impact assessment for Large Language Models (LLMs) within a foundation-model paradigm. It outlines the Act's phased implementation, broad AI-system definition, and high-risk classification framework, and examines how LLMs fit into this regime. It argues that LLMs pose risks such as hallucinations, bias, and privacy concerns that warrant human oversight, ethical governance, and DPIAs, even when not automatically classified as high-risk. It also surveys ongoing liability proposals and emphasizes the need for integrated governance to safeguard rights and ensure responsible deployment in professional contexts.

Abstract

The imposing evolution of artificial intelligence systems and, specifically, of Large Language Models (LLM) makes it necessary to carry out assessments of their level of risk and the impact they may have in the area of privacy, personal data protection and at an ethical level, especially on the weakest and most vulnerable. This contribution addresses human oversight, ethical oversight, and privacy impact assessment.