Civil liability of platforms for AI-generated content

1. The fact

   The issue of civil liability for erroneous outputs emerged strongly in Italy following the interventions of the Italian Data Protection Authority, which, in March 2023, raised the issue of data inaccuracy, highlighting how algorithmic hallucinations can harm the reputation of the individuals involved.

   As reported by Il Sole 24 Ore, the debate has focused on the need to ensure the accuracy of information provided by language models. Currently, the matter is in a phase of defining jurisdiction and the scope of liability: while there are no definitive Supreme Court rulings yet, the procedural debate is focused on qualifying the provider as an active hosting provider. The established facts show that generative models can produce false statements presented as factual truths, prompting the injured parties' legal counsel to challenge the immunity typically enjoyed by digital intermediaries.

2. The rules at play

   The regulatory framework relies on Legislative Decree 70/2003 and the Digital Services Act (EU Regulation 2022/2065), which govern the liability exemption for intermediary service providers. However, Civil Code provisions play a crucial role: Art. 2043 c.c. for general tort liability, Art. 2049 c.c. for the liability of masters and employers applied by viewing the algorithm as an auxiliary, and above all Art. 2050 c.c. regarding the exercise of dangerous activities.

   The latter provision is invoked because the deployment of generative AI without verification filters could be equated to an inherently risky activity, shifting the burden of proof to the provider.

3. What the case law says

   Case law from higher courts has clarified the distinction between a passive hosting provider and an active one over time. According to established jurisprudence, a provider that indexes, organizes, or generates content loses the benefit of liability exemption if it fails to prove that it acted with the required degree of diligence.

   The Court of Justice of the EU has likewise emphasized that immunity is forfeited when the platform operator plays an active role that affords it knowledge of or control over the data. Applying these principles to AI, the legal debate explores the possibility of holding software companies directly liable for generated output, assessing their position regarding any resulting reputational damage.

5. Analysis drafted and verified with edit.legal

   To verify the provisions cited in this article, we used edit.legal. Test our legal AI on official sources and apply it to your own matters.

   Try edit.legal AI

8. Lessons for professionals

   The analysis of the legal and regulatory framework yields crucial takeaways for legal practice. First, securing and preserving technical evidence of the harmful output before the model is updated or modified is paramount. Second, it highlights the importance of pleading concurrent heads of liability, ranging from general tort liability under Art. 2043 c.c. to strict liability for dangerous activities under Art. 2050 c.c.

   Third, issuing a timely notification to the provider is a key step to establish the intermediary's actual knowledge of the infringement for the purposes of the Digital Services Act.

References: Art. 2043 Codice CivileArt. 2049 Codice CivileArt. 2050 Codice CivileD.lgs. 70/2003Regolamento (UE) 2022/2065 (Digital Services Act)

Editorial oversight: Avv. Federico Papa·ICAMUpdated 27 May 2026