A (in)existência do dever legal de veracidade na IA generativa: uma análise crítica do PL 2.338/2023

Abstract

This article presents a critical analysis of the performance of generative artificial intelligence systems, especially large language models (LLMs), focusing on the risks arising from the dissemination of automated discourses lacking truthfulness and epistemic commitment. Based on the discussion of so-called careless speech, it was identified that these systems, despite their high textual fluency, tend to produce biased, inaccurate, or misleading content, undermining the reliability of the informational ecosystem and threatening fundamental rights. In light of the insufficiency of corporate self-regulation strategies, the need for state intervention through effective normative mechanisms is addressed. In this context, na analysis was conducted of Draft Bill no. 2.338/2023, which seeks to establish the Legal Framework for Artificial Intelligence in Brazil, aiming to verify whether the legislative proposal contains mechanisms capable of imposing a duty of truthfulness on generative AIs. It concludes that, although the bill introduces relevant principles – such as information integrity and system reliability – it lacks binding provisions that ensure a concrete obligation of truthfulness on the part of such systems. The absence of specific regulation on this issue weakens the legal protection of users and poses challenges to the consolidation of a plural, safe, and democratically aligned informational environment.