This short article will focus on the most intrusive, pervasive, transforming and revolutionary change that has occurred in recent years and has (and will) severely impact the legal profession. I am, of course, discussing Artificial Intelligence (AI).
Lawyers of my generation have come all the way down from holding a pen to using a word processor, connecting to the first digital databases of the Court of Cassation, exchanging emails and documents via cell phones and using all the other technological devices of our current world.
In all those cases, we have used tangible and intangible “tools”, i.e. assets which were possessed and directed by a lawyer; in one word, things that the lawyer uses. My perception is that the situation has radically changed.
- What is AI really?
- She moves in mysterious ways. AI is self-learning and autonomous[1], That means that AI is not a (static, passive) “tool”. A tool is, in fact, something that needs to be activated by human push and guidance. Hence, it can be possessed. I own a car because it remains still unless I switch on the engine and drive it. Moreover, the car moves when I steer it. The same happens with software: a lawyer can instruct it, and the software produces exactly what it is programmed to do.
AI is not that. AI is dynamic, active and independent. AI explores data, finds (hidden, surprising) patterns, auto-learns by experience, evolves, and provides outputs which go beyond the inputs[2]. As a consequence, the relationship between the human user and the IT artefact is not that between an active individual (owner) and a subject matter (an owned piece of nature). In the latter case, ownership can be a paradigm, inasmuch as it reflects the asset’s subjection to the owner. When we discuss AI, that cannot be the paradigm. One cannot own AI. A tool is handled and used by the human being (the professional), while AI is consulted. The relationship between human beings and technology is at an entirely different level. Hence, we need to understand the transformative nature of AI and adapt our legal reading to the new situation.
- A shift of power. Another footprint of AI is its changing power infrastructure [3]. The classical power allocation relies on the concept of discretion. We have gone from the absolute discretion of the monarch to the reasonable discretion of the civil servant (with a view to the public good)[4] or the business judgment of the directors in a corporation (with a view to achieving the corporate benefit)[5]. Yet discretion remains discretionary, and the judiciary cannot challenge the merits of the decision but only the process by which it was made.
Something has changed with the AI, though. For the first time in history, one can scrutinise the exercise of human discretion by using technology. Indeed, by computing and analysing millions or billions of recorded data points, AI can determine, statistically and based on the context provided to the machine, whether the decision was likely to be right or wrong at the time it was taken. The choice of a civil servant (e.g., when approving civil works), a director (e.g., when deciding on a transaction), or a lawyer (e.g., when establishing a defence strategy) can be benchmarked against the output of a technology.
An enormous shift of power is therefore taking place. The concept of human discretion, to a certain extent non-challengeable, is disrupted by the emergence of a competitive species: AI[6].
- Threats and false myths
- My (professional) kingdom for a horse. The change comes with opportunities and threats, including in the legal profession.
The opportunities for the practitioner certainly lie in the ability to analyse a large volume of information, broaden and deepen the legal search, and facilitate the reduction of legal thinking in legal drafting. On top of that, there is an effect that benefits society as a whole that I would call “legal devolution”, i.e., the diffusion of legal concepts and solutions to the public by societal players in a simple and accessible manner, thereby cutting agency costs.
On the other hand, there are at least two main threats. In the first instance, while it is reasonable to expect that so-called hallucinations will be progressively reduced with improvements in technology and the supply of more reliable context to the machine, I believe that a serious threat is posed by conformity, which arises as a by-product of AI. Artificial Intelligence provides so many relaxing and well-presented summaries, tables, and recommendations that one can forget that these outputs are nothing but the result of a statistical recapitulation of the past. As a matter of fact, AI does not reason but only puts together bits and pieces. More, AI can only recompile (in a very well-ordered fashion) the already existing interpretations of law, the standard ways certain clauses are drafted, or certain defensive arguments have been attempted in Court. In other words, legal AI outputs do not bring anything new. Relying on AI, thus, jeopardises evolution[7].
In the second instance, surrendering information to AI providers can put legal privilege at risk because AI providers are not lawyers and process the information in other Countries, subject to their legislation and governments [8].
- Lullaby. Confronted with the complexities of AI (and the above threats), the construction of a (false) myth began, that AI may be (legislatively, judicially) downsized and reduced to a customary tool on which the professional can prevail (command) in the frame of the so-called human-centrism[9]. The guidelines issued by the law bars, the decisions of the Courts and, likewise, the recent rules enacted in Italy go in that direction, in as much as they impose that lawyers should use AI only for support, ancillary functions (as if AI were a traditional technological device) and, in any event, lawyers must re-check the AI outputs, one by one, from inception to end[10].
However, cutting the wings to AI does not seem like a good idea. Moreover, it does not seem like a doable approach: in fact, which authority can claim a legal mandate to scrutinise whether a lawyer has used AI only for support or more than that? And how is the scrutiny intended to be carried out? Are we really willing to have governmental agencies, or even our Bar, step into our offices and criticise how we use our technology to defend our clients?[11] In my eyes, dictating the obligation of a lawyer to use AI in one way or the other will not produce beneficial results and all the same applies if we interpret the law in the sense that lawyers have to recheck the work-product of AI. There is indeed an alternative path that we could follow.
- I’m only human. The alternative I would like to present to your reflection is based on two pillars.
First, human oversight (as required under both the AI Act and the Italian AI law) should be interpreted as applying to the workflow, not to the work product [12]. In other words, the legal supervision duty does not entail checking each AI output and redoing the exercise already completed by the AI. Human oversight applies to the work system in which the AI query is one (but not the only) element/step in the process. In the light of that, supervising the system does not imply human correction of the AI alleged mistakes(which would be impossible because of the reduced computation abilities of human beings as compared to AI models), but to check that the elements forming the work-flow are working as programmed, i.e.: the data sets are appropriately fed, available and updated, the maintenance is regularly carried out, the IT connections and interfaces work, there are no systemic anomalies, there have not been incidents, data breaches or hacking/data poisoning. In this sense, human oversight is a second-level control, indeed a check on the organisation, not on the inner workings of the software.
Secondly, human-centrism should be read as humanism. In that conceptual framework, human-centrism means that AI must be directed towards the benefit of humanity[13]. Therefore, AI should be programmed, trained, marketed and used in accordance with our societal values[14] and moral rules[15]. To that purpose, it is unnecessary to predicate that humans command the AI. What is crucial is that we combine AI energies with human objectives[16].
The result is that human intervention must apply in conjunction with the automated output, not to replace it but to exercise governance on the process and, eventually, to complement it with the human viewpoint, which is based on a variety of non-computational and non-statistical abilities rooted in the creative force of law interpretation, legal construction, and moral values.
3. The practice of law and the law in practice
So what? For all those reasons, I think that the emphasis should shift from pretending to control the AI’s output (the product) to controlling the AI system (the process)[17]. In the specific case of lawyers, there are a couple of things that law firms may truly do very well:
-
- Create a community of AI-conscious professionals[18]. That entails combining solid legal skills with AI abilities at every level of seniority. Only those who use AI can develop experience and awareness[19]. Only senior lawyers who know how to use AI can guide and, if necessary, correct the work of junior lawyers who have used it.
- Negotiate AI contracts effectively. Law firms should insist on certifications, credentials, warranties and insurance coverage by their providers[20]. Lawyers should avoid technological lock-in[21], rotate providers. Most importantly, they have to defend secrecy[22] and cybersecurity, keep title to the inputs and the outputs. EU regulations, including the Data Act, offer an interesting range of legal protections
4. Conclusions: dancing in the dark. After all, optimistically, one can think that the legal profession will be resilient. Lawyers’ natural effort to find new ways to defend their clients will push them to remain creative and explore new pathways. More than that, lawyers are best placed to stay attuned to the spirit of the law and the embedded values of our society. In fact, that is precisely the role that I think should be carried out by human intervention and the reason why decisions cannot be taken solely through automated means[23]. Lawyers have to continue to benchmark the AI outputs with legal values. Fairness, equity, trust, loyalty, and fiduciary duties cannot be reduced to automated empirical statistics[24]. Interpretation of laws does not equate to mere recollection of existing rules, nor to a probabilistic summary of the order of words used on the Internet.
[1] Art.3(1) of the EU Reg. 1689 of 13 June 2024 (the AI Act): “AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.”. In the same sense, see whereas 12 of the AI Act: “The adaptiveness that an AI system could exhibit after deployment, refers to self-learning capabilities, allowing the system to change while in use.” For a very good analysis of AI, see N. Abriani, Intelligenza artificiale e diritto societario, in Enc. Dir., I tematici, IX, 2025. It is also meaningful that AI is legally defined either as a “model” or a “system” (not as a “tool”), where a model is a set of ideas and numbers describing the past, present, or future state of something and a system is a portion of the universe consisting of related elements working together. Those terms go far beyond a mere asset, thing or tool.
[2] Again, whereas 12 of the AI Act: “A key characteristic of AI systems is their capability to infer. This capability to infer refers to the process of obtaining the outputs, such as predictions, content, recommendations, or decisions, which can influence physical and virtual environments, and to the capability of AI systems to derive models or algorithms, or both, from inputs or data. The techniques that enable inference in AI systems include machine learning approaches that learn from data to achieve certain objectives, and logic- and knowledge-based approaches that infer from encoded knowledge or symbolic representations of the task to be solved. The capacity of an AI system to infer transcends basic data processing by enabling learning, reasoning or modelling.” P. Benanti, Siamo arrivati alla resa: quando l’IA ha smesso di essere un mezzo, https://paolo1973.substack.com/p/siamo-arrivati-alla-resa-quando-lia?r=5637y9&utm_medium=ios&utm_source=notes-share-action.
[3] O. Pollicino, L’intelligenza artificiale non è più uno strumento: è architettura del potere, https://www.agendadigitale.eu/cultura-digitale/lintelligenza-artificiale-non-e-piu-uno-strumento-e-architettura-del-potere/. You can find many stimulating thoughts about the relation between power and discretion in the writings of Michel Foucault.
[4] M. Nigro, Giustizia Amministrativa, Il Mulino, 2002.
[5] https://giurisprudenzadelleimprese.it/sentenze/?_search=business%20judgement%20rule
[6] It is not surprising that some has therefore opined that directors or lawyers have a fiduciary duty to consult AI before making choices. A.M. Gambino, La responsabilità dell’Avvocato per mancato utilizzo della IA, https://dirittodiinternet.it/wp-content/uploads/2025/12/relazione.pdf; E. Rimini, Intelligenza artificiale e doveri degli amministratori, https://www.rivistacorporategovernance.it/Article/Archive/index_html?ida=161&idn=18&idi=-1&idu=-1.
Lost or reduced discretion equates to lost or reduced power. In a way, something is happening which resembles alienation, which was first theorised by Marx when he observed that manual workers manufactured products that were not theirs but of the capitalists. Now, with the emergence of AI, it seems that entrepreneurs are partially stripped of their traditional function of making decisions on the management and organisation of the enterprise. The power shifts in new hands, those who hold the technology and may provide it to the business undertakings.
[7] As far as law (and the rule of law) is concerned, one of the main risks is that lawyers and judges too heavily rely on AI, so that they become in effect human interfaces for AI decisions. That would be a radical danger. AI provides predictions based on statistical analysis of available data. Therefore, AI cannot create, diverge, or overturn a legal point. It will simply reflect the predominance of traditional interpretations of the past. Uniformity and conformity may switch off the enlightenment and hypnotise innovation.
[8] https://www.ibanet.org/Digital-strangers-in-litigation; https://jolt.law.harvard.edu/digest/against-an-ai-privilege; see United States v Heppner (Judge Jed Rakoff of the Southern District of New York, 17 February 2026) and Warner v Gilbarco Inc (Magistrate Judge Anthony P. Patti of the Eastern District of Michigan, 10 February 2026).
[9] The term “human-centrism” is rather ambiguous. My viewpoint is that we should not interpret it as if humans were (should be) the centre of the universe. That would echo old visions where humans originate from God and therefore have an inner divine nature. One could rightly remark that those who sentenced Galileo to prison did so based on human-centrism (Brecht, Life of Galileo). Anyway, the real risk of that interpretation is that we delude ourselves into thinking that we can change reality through legal rules. There is a large body of literature claiming that AI can do things humans cannot. There is thus a sort of psychological removal in claiming that humans shall command something they cannot understand. Too large is the asymmetry in computational abilities between individuals and AI.
[10] Council of Bars and Law Societies of Europe, Guide on the use of generative AI by lawyers, 2 October 2025, https://www.ccbe.eu/fileadmin/speciality_distribution/public/documents/IT_LAW/ITL_Guides_recommendations/EN_ITL_20251002_CCBE-guide-on-the-use-of-the-use-of-generative-AI-for-lawyers.pdf: “To avoid or mitigate the risks posed by the use of GenAI tools in legal practice, lawyers should verify the output of a GenAI before it is utilised in their work (where the use case requires), understand the capabilities and limitations of all technological solutions they use for their work, including GenAI and understand the different contexts in which they use GenAI as well as the implications and risks that arise from such use.” Syracuse Tribunal, 20 February 2026, n. 338: lawyers may GenAI diligently verifying the reliability of the output and consulting the quoted case-law directly; failing that, there is gross negligence on their part (commented by G. Cassano, Avvocati e AI generative: doverosa la verifica dell’attendibilità degli output, Dir. Internet, 2026, 331). Art. 13 Italian Law 25 September 2025 no. 132: use of AI in the profession is only aimed at instrumental and support activities, where the intellectual exercise by the professional must prevail. Circular issued by the Italian National Bar of Lawyers (CNF) on 13 October 2025 (https://www.consiglionazionaleforense.it/comunicazioni-per-gli-ordini/-/asset_publisher/Snwi9zTL8Vut/content/id/3762821/pop_up?_com_liferay_asset_publisher_web_portlet_AssetPublisherPortlet_INSTANCE_Snwi9zTL8Vut_viewMode=print&_com_liferay_asset_publisher_web_portlet_AssetPublisherPortlet_INSTANCE_Snwi9zTL8Vut_languageId=it_IT) states that the AI shall only be used for support functions to the profession, such as, without limitation, organizational and secretary activities, legal research, preliminary analysis of documents and compilation of drafts and summaries, provided that the output shall be carefully and accurately verified by the engaged lawyer, as in the generation of the document as in the check of legal precedents or rules. In the same sense the Vademecum per avvocati sull’uso dell’I.A., Commissione procedura civile dell’Ordine degli Avvocati di Roma, 2026. See also F. Paolucci, M. Bassini, O. Pollicino, L’imperativo: verifica dei risultati, NT+ Diritto, 18 June 2025. G. Finocchiaro, L’intelligenza artificiale in ambito giudiziario, Riv. Trim. Dir. Proc. Civ., 2024, p. 425.
[11] There is a widely diffused literature on the ethical duty of lawyers to accept only “support” by AI but firmly refuse the replacement of their profession by automated outputs. It is hard to conceive who will check that. Most importantly, I do not see in practice how that may be enforced. The borderline is blurred. How a Law Bar or a Court cand determine whether a lawyer has been (just) helped or (actually) substituted in forming a legal opinion? Are we intending that the Bar will scrutinise how the lawyer has acted? Are we saying that the lawyer should simply change a word or two to overcome the test? Or are we saying that lawyers should redo the same exercise as that of the AI and then compare their human output to the automated output? If so, why should I use the AI in the first place?
[12] It appears that the AI Act discusses human oversight (Art. 14), not human command. Art.1.3 of Italian Law 25 September 2025 no. 132 (hereinafter, the Italian AI Law) establishes that AI systems and models must be developed and applied under human oversight and remain subject to human intervention.
[13] That is the sense to be attributed, in my view, to Art. 1.1 of the Italian AI Law whose objective is to promote “a correct use of AI in a human-centric dimension”. Art. 19.2 of same law indicates that the AI national strategy will take into account human rights. See also whereas 6 of the AI Act: “As a prerequisite, AI should be a human-centric technology. It should serve as a tool for people, with the ultimate aim of increasing human well-being. See Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, signed in Vilnius on 5 November 2024 (https://rm.coe.int/1680afae3c).
[14] P. Pollice, Il contratto (appunti per un corso di diritto civile), Turin, 2015, p. 132-133.
[15] G. Rossi, Capitalismo e diritti umani, Riv. Dir. Soc., 2011, p. 3. P. Gianniti, Il faticoso procedere della persona umana tra i crinali della storia, Dir. Fam. Pers., 2021, 9. 1243. The goal is not that, through AI, few can become more than human, but that mankind becomes more human. According to G. Calamandrei, http://www.cristinacampo.it/public/discorso%2, laws are fluxes of reasoning which adapt to the real world.
[16] See also https://digital-strategy.ec.europa.eu/en/news/commission-publishes-guidelines-providers-general-purpose-ai-models.
[17] That is the spirit of Whereas 71 of EU reg.679 of 27 April 2016 no. 679 (GDPR), where the data controller is mandated to implement suitable measures which include: the use of appropriate mathematical or statistical procedures, technical and organizational measures to ensure correction of inaccuracies, and risk of errors minimized, secure personal data. See also, as far as corporate governance is concerned, M. Rescigno, Note in tema di responsabilità degli emittenti, Competitività dei capitali e riforma del testo unico della finanza, Quad. Giur. Comm., 2025, applies a similar approach to the interaction between directors of corporations and AI.
[18] Art. 4 of the AI Act provides for a duty to guarantee the so-called AI literacy.
[19] G. Vaciago, Intelligenza artificiale generativa e professione forense, Milan, 2024. A.M. Gambino e D. Russo, Venti errori degli avvocati che usano l’IA, Dir. Mer. Tecn., 2026.
[20] The Milan Bar has started the project of Horos Hub, which is a valid answer to the need (https://www.ordineavvocatimilano.it/it/mappatura-delle-soluzioni-di-intelligenza-artificiale-horos-e-losservatorio-ai-dellordine-degli-avvocati-di-milano/p719). G. Ziccardi, Intelligenza artificiale guida semplice per il giurista del futuro, Chapter IV, Milan, 2025.
[21] See the provisions of UE Regulation 2854 of 13 December 2023 (Data Act) and the Commission Recommendation on non-binding model contractual terms on data access and use and non-binding standard contractual clauses for cloud computing contracts (https://digital-strategy.ec.europa.eu/en/library/draft-recommendation-non-binding-model-contractual-terms-data-access-and-use-and-non-binding). R. Grisafi, La portabilità dei dati nella prospettiva del Data Act, Giur. It., 2026, p. 490.
[22] Art. 28.3 of the Code of Ethics of Italian Lawyers: “L’avvocato deve adoperarsi affinché il rispetto del segreto professionale e del massimo riserbo sia osservato anche da dipendenti, praticanti, consulenti e collaboratori, anche occasionali, in relazione a fatti e circostanze apprese nella loro qualità o per effetto dell’attività svolta.” Same duty is entrenched in CCBE Model Code of Conduct. A. Bonifati, Sul Segreto professionale dell’avvocato, Riv. Prev. For., 2025.
[23] Art. 22 of the GDPR. Art. 15 of the Italian AI Law.
[24] G. D’Acquisto, Decisioni algoritmiche, Turin, 2022.
