This research stems from the observation that artificial intelligence systems are becoming increasingly prevalent in key areas of social life and, consequently, from the need to assess their impact on fundamental aspects of criminal law. In particular, the study examines whether traditional models of attributing liability are still adequate for addressing phenomena characterised by algorithmic decision-making autonomy, or whether a systematic rethinking of these models is required. The study focuses on the central issue of the causal link, which is called into question by the opacity and unpredictability inherent in machine learning systems (so-called ‘black boxes’). In this context, it was deemed appropriate to emphasise the criterion of logical probability as a tool for assessment in situations of uncertainty, whilst proposing a reinterpretation of causality in ‘psychic’ terms, suitable for describing the relationship between human input and the system’s autonomous output. From this perspective, the action of the human agent is not understood as a deterministic cause of the event, but rather as a factor guiding a decision-making process which, whilst not fully predictable, remains anchored to the initial conditions set by the human agent. As regards culpability, the research highlights the risk of drifting towards forms of liability detached from the facts, suggesting instead a model centred on the management of technological risk. In this light, Regulation (EU) 2024/1689 (AI Act) is considered not only as an instrument of administrative regulation, but also as a possible source of precautionary rules relevant to the assessment of culpability. Of particular importance in this context are the concepts of culpa in istruendo and culpa in vigilando, as well as the need for effective and meaningful human oversight of intelligent systems. The research also rules out the viability of theories proposing the recognition of autonomous legal personality for artificial intelligence systems, considering it necessary to uphold the centrality of the principle of humanity in punishment. In the business context, it has been considered that the provisions of Legislative Decree No. 231/2001 represent the most appropriate instrument for the attribution of liability, emphasising the category of organisational negligence in contexts characterised by complex technological risks. The work is structured around three main strands: a first part dedicated to the technical and functional overview of the phenomenon of artificial intelligence; a second part centred on the analysis of attribution models, with particular reference to the profiles of causality and culpability in the light of European legislation; and a third part, finally, aimed at verifying the validity of the categories examined in certain high-risk sectors, such as surgical robotics and autonomous driving. The overall objective is to contribute to the development of a model of criminal law for artificial intelligence capable of combining the effective protection of legal interests with respect for fundamental human rights

La presente ricerca prende le mosse dalla constatazione della crescente incidenza dei sistemi di intelligenza artificiale nei principali ambiti della vita sociale e, conseguentemente, dalla necessità di verificarne l’impatto sulle categorie fondamentali del diritto penale. In particolare, il lavoro si interroga se i modelli tradizionali di imputazione della responsabilità siano ancora idonei a fronteggiare fenomeni caratterizzati da autonomia decisionale algoritmica, ovvero se sia necessario un loro ripensamento in chiave sistematica. L’indagine si concentra sul nodo centrale del nesso di causalità, messo in crisi dall’opacità e dall’imprevedibilità proprie dei sistemi di machine learning (c.d. black box). In tale contesto, si è ritenuto di valorizzare il criterio della probabilità logica quale strumento di accertamento in situazioni di incertezza, proponendo al contempo una rilettura della causalità in termini “psichici”, idonea a descrivere il rapporto tra input umano e output autonomo del sistema. In questa prospettiva, l’azione dell’agente umano non viene intesa come causa deterministica dell’evento, bensì come fattore che orienta un processo decisionale che, pur non essendo pienamente prevedibile, resta ancorato alle condizioni iniziali poste dall’uomo. Quanto al versante della colpevolezza, la ricerca evidenzia il rischio di derive verso forme di responsabilità sganciate dal fatto, suggerendo invece un modello incentrato sulla gestione del rischio tecnologico. In tale ottica, il Regolamento (UE) 2024/1689 (AI Act) viene considerato non soltanto quale strumento di regolazione amministrativa, ma anche come possibile fonte di regole cautelari rilevanti ai fini del giudizio di rimproverabilità. Particolare rilievo assumono, in questa prospettiva, le figure della culpa in istruendo e della culpa in vigilando, nonché l’esigenza di un controllo umano effettivo e significativo sui sistemi intelligenti. La ricerca esclude, inoltre, la percorribilità delle teorie che prospettano il riconoscimento di una soggettività giuridica autonoma in capo ai sistemi di intelligenza artificiale, ritenendo necessario mantenere ferma la centralità del principio di umanità della pena. In ambito imprenditoriale, si è ritenuto che la disciplina di cui al d.lgs. n. 231/2001 rappresenti lo strumento più adeguato per l’imputazione della responsabilità, valorizzando la categoria della colpa di organizzazione nei contesti caratterizzati da rischi tecnologici complessi. Il lavoro si articola lungo tre direttrici principali: una prima parte dedicata alla ricognizione tecnico-funzionale del fenomeno dell’intelligenza artificiale; una seconda incentrata sull’analisi dei modelli imputazionali, con particolare riferimento ai profili di causalità e colpevolezza alla luce della normativa europea; una terza, infine, volta a verificare la tenuta delle categorie esaminate in alcuni settori ad alto rischio, quali la robotica chirurgica e la guida autonoma. L’obiettivo complessivo è quello di contribuire alla costruzione di un modello di diritto penale dell’intelligenza artificiale capace di coniugare l’effettività della tutela dei beni giuridici con il rispetto delle garanzie fondamentali della persona.

Intelligenza artificiale e diritto penale: nuove soluzioni per nuovi problemi o vecchie soluzioni per vecchi problemi? / Battaglia, A.. - (2026 Oct 21).

Intelligenza artificiale e diritto penale: nuove soluzioni per nuovi problemi o vecchie soluzioni per vecchi problemi?

Battaglia, Andrea
2026-10-21

Abstract

This research stems from the observation that artificial intelligence systems are becoming increasingly prevalent in key areas of social life and, consequently, from the need to assess their impact on fundamental aspects of criminal law. In particular, the study examines whether traditional models of attributing liability are still adequate for addressing phenomena characterised by algorithmic decision-making autonomy, or whether a systematic rethinking of these models is required. The study focuses on the central issue of the causal link, which is called into question by the opacity and unpredictability inherent in machine learning systems (so-called ‘black boxes’). In this context, it was deemed appropriate to emphasise the criterion of logical probability as a tool for assessment in situations of uncertainty, whilst proposing a reinterpretation of causality in ‘psychic’ terms, suitable for describing the relationship between human input and the system’s autonomous output. From this perspective, the action of the human agent is not understood as a deterministic cause of the event, but rather as a factor guiding a decision-making process which, whilst not fully predictable, remains anchored to the initial conditions set by the human agent. As regards culpability, the research highlights the risk of drifting towards forms of liability detached from the facts, suggesting instead a model centred on the management of technological risk. In this light, Regulation (EU) 2024/1689 (AI Act) is considered not only as an instrument of administrative regulation, but also as a possible source of precautionary rules relevant to the assessment of culpability. Of particular importance in this context are the concepts of culpa in istruendo and culpa in vigilando, as well as the need for effective and meaningful human oversight of intelligent systems. The research also rules out the viability of theories proposing the recognition of autonomous legal personality for artificial intelligence systems, considering it necessary to uphold the centrality of the principle of humanity in punishment. In the business context, it has been considered that the provisions of Legislative Decree No. 231/2001 represent the most appropriate instrument for the attribution of liability, emphasising the category of organisational negligence in contexts characterised by complex technological risks. The work is structured around three main strands: a first part dedicated to the technical and functional overview of the phenomenon of artificial intelligence; a second part centred on the analysis of attribution models, with particular reference to the profiles of causality and culpability in the light of European legislation; and a third part, finally, aimed at verifying the validity of the categories examined in certain high-risk sectors, such as surgical robotics and autonomous driving. The overall objective is to contribute to the development of a model of criminal law for artificial intelligence capable of combining the effective protection of legal interests with respect for fundamental human rights
21-ott-2026
Settore IUS/17 - DIRITTO PENALE
Settore GIUR-14/A - Diritto penale
Squillaci, Ettore Michele
TROPEA, Giuseppe
Doctoral Thesis
File in questo prodotto:
File Dimensione Formato  
PhD Thesis_Battaglia Andrea_XXXVIII ciclo.pdf

accesso aperto

Tipologia: Tesi di dottorato
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 3.19 MB
Formato Adobe PDF
3.19 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12318/170366
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact