1 00:00:04,820 --> 00:00:05,480 NICOLA FABIANO 2 00:00:05,480 --> 00:00:05,599 ITALIAN LAWYER 3 00:00:08,220 --> 00:00:09,539 NICOLA FABIANO 4 00:00:09,000 --> 00:00:13,020 Hello, I'm Nicola Fabiano, an Italian lawyer, 5 00:00:09,600 --> 00:00:09,619 NICOLA FABIANO 6 00:00:13,260 --> 00:00:18,780 and welcome to the first episode of a new series dedicated to legal prompting. 7 00:00:18,920 --> 00:00:23,659 Before we dive in, I want to clarify one point straight away. 8 00:00:23,920 --> 00:00:27,879 This series is not meant to explain, in general terms, 9 00:00:27,879 --> 00:00:33,180 how to use chatGPT or, more broadly, generative AI systems. 10 00:00:33,419 --> 00:00:38,880 It is not a tutorial, nor is it meant to follow the kind of superficial enthusiasm 11 00:00:39,439 --> 00:00:41,919 that often surrounds these tools. 12 00:00:42,000 --> 00:00:46,299 The goal is different, to understand how artificial intelligence 13 00:00:46,299 --> 00:00:53,099 can be used in legal and professional contexts with method, awareness, and critical judgment. 14 00:00:53,099 --> 00:01:00,439 When I speak of legal prompting, I do not simply mean the ability to ask questions to a language model. 15 00:01:00,680 --> 00:01:05,739 I mean the ability to formulate instructions in a legally structured way, 16 00:01:06,220 --> 00:01:11,260 taking into account the logic of legal reasoning, the hierarchy of legal sources, 17 00:01:11,699 --> 00:01:16,180 interpretive criteria, professional duties, and the practical consequences 18 00:01:16,180 --> 00:01:21,959 that a response may have for a client, an organization, or an operational decision. 19 00:01:21,959 --> 00:01:24,879 And this is where the real difference lies. 20 00:01:25,260 --> 00:01:30,199 The difference between a generic prompt and a well-constructed legal prompt 21 00:01:30,199 --> 00:01:34,180 does not lie in the technology. It lies in the method. 22 00:01:34,459 --> 00:01:37,400 And, in law, method cannot be improvised. 23 00:01:37,699 --> 00:01:42,819 A legal professional cannot simply accept a text because it appears plausible. 24 00:01:43,139 --> 00:01:50,080 It must be evaluated, verified, contextualized, and, where necessary, substantially revised. 25 00:01:50,080 --> 00:01:56,980 In other words, you cannot delegate to a model what you are not able to assess critically yourself. 26 00:01:57,260 --> 00:02:00,480 Three fundamental premises will guide this series. 27 00:02:00,779 --> 00:02:04,639 The first is that language models do not reason like lawyers. 28 00:02:05,099 --> 00:02:10,440 They produce plausible text, often well-written, sometimes even persuasive. 29 00:02:10,699 --> 00:02:14,380 But plausibility is not the same as legal accuracy. 30 00:02:14,380 --> 00:02:19,380 A model may confuse legal sources, cite provisions that do not exist, 31 00:02:20,440 --> 00:02:25,979 mix different jurisdictions, invert the relationship between general and special rules, 32 00:02:26,179 --> 00:02:30,960 or produce summaries that are formally elegant but substantively wrong. 33 00:02:31,199 --> 00:02:36,039 For that reason, human oversight is not optional. It is central. 34 00:02:36,300 --> 00:02:42,520 And for those working in regulated professions, it is also a specific professional responsibility. 35 00:02:42,520 --> 00:02:50,479 The second premise is that the use of artificial intelligence in professional settings does not take place in a vacuum. 36 00:02:50,740 --> 00:02:55,919 There is already a legal and regulatory framework that requires close attention. 37 00:02:56,259 --> 00:03:04,199 The GDPR, the AI Act, the principles of confidentiality, data minimization, and accountability, 38 00:03:04,199 --> 00:03:12,419 professional ethical duties, and the organizational obligations that apply to law firms, companies, and public authorities. 39 00:03:12,699 --> 00:03:17,199 That means AI is not a neutral tool to be adopted lightly. 40 00:03:17,479 --> 00:03:24,960 It is a powerful tool, but also a regulated one, and its use requires informed and responsible choices. 41 00:03:25,279 --> 00:03:28,679 The third premise is perhaps the most underestimated. 42 00:03:28,679 --> 00:03:33,880 The choice of model and infrastructure is not a purely technical decision. 43 00:03:34,279 --> 00:03:38,419 It is also, and often above all, a compliance decision. 44 00:03:38,740 --> 00:03:47,320 Using a cloud-based system to process confidential information, personal data, or material protected by professional secrecy, 45 00:03:47,539 --> 00:03:52,660 without having properly assessed the relevant legal and organizational requirements, 46 00:03:52,899 --> 00:03:57,380 can expose professionals and organizations to very serious risks. 47 00:03:57,380 --> 00:04:03,380 We must ask where the data goes, who processes it, for what purposes, with what safeguards, 48 00:04:04,399 --> 00:04:11,520 under which contractual terms, and with what implications in terms of transfers, security, and liability. 49 00:04:11,639 --> 00:04:16,859 In other words, even before asking what can this model do, we should ask, 50 00:04:17,179 --> 00:04:21,119 under what conditions can I use it lawfully and responsibly? 51 00:04:21,399 --> 00:04:24,640 And that is exactly why this series was created. 52 00:04:24,640 --> 00:04:31,859 In recent months, legal prompting has become increasingly central to my work, research, and training activities. 53 00:04:32,179 --> 00:04:38,619 I address it in articles, public speaking engagements, academic contexts, and professional discussions. 54 00:04:38,859 --> 00:04:45,820 But I wanted to create a space that is more direct, more reflective, and, in a sense, more practical. 55 00:04:46,160 --> 00:04:49,200 This podcast is meant to be exactly that. 56 00:04:49,200 --> 00:04:55,559 A space for in-depth discussion where we can address concrete cases, real-world problems, 57 00:04:55,820 --> 00:05:01,940 and the questions faced by those working in law, compliance, data protection, cybersecurity, 58 00:05:02,440 --> 00:05:05,119 and, more broadly, innovation governance. 59 00:05:05,239 --> 00:05:10,000 So we will not be discussing artificial intelligence in abstract terms. 60 00:05:10,279 --> 00:05:15,700 We will discuss how to use it and how not to use it when drafting a privacy notice, 61 00:05:15,700 --> 00:05:19,160 analyzing a decision of a data protection authority, 62 00:05:19,399 --> 00:05:21,019 preparing a legal assessment, 63 00:05:21,420 --> 00:05:24,700 designing a retrieval system or a legal AI project, 64 00:05:25,480 --> 00:05:30,940 or integrating AI into the internal processes of a law firm or an organization, 65 00:05:31,320 --> 00:05:33,119 without creating new risks. 66 00:05:33,420 --> 00:05:36,119 We will also focus on another essential point. 67 00:05:36,380 --> 00:05:40,619 Using these tools well does not simply mean getting better answers. 68 00:05:40,940 --> 00:05:43,660 Above all, it means building better processes. 69 00:05:43,660 --> 00:05:47,200 Because a good prompt on its own is not enough. 70 00:05:47,480 --> 00:05:51,160 What is needed is the right context, a method of verification, 71 00:05:51,600 --> 00:05:55,100 and the ability to distinguish between operational support 72 00:05:55,100 --> 00:05:58,760 and the improper substitution of professional judgment. 73 00:05:59,119 --> 00:06:04,880 That is why, at least as I understand it, legal prompting is not an isolated technique. 74 00:06:05,119 --> 00:06:09,820 It is a point of convergence between legal expertise, critical judgment, 75 00:06:10,160 --> 00:06:13,059 data governance, and professional responsibility. 76 00:06:13,059 --> 00:06:16,880 In the coming episodes, we will go further into detail. 77 00:06:17,160 --> 00:06:20,959 We will discuss the structure of legal prompts, common mistakes, 78 00:06:21,279 --> 00:06:24,679 the reliability of outputs, sources and controls, 79 00:06:24,959 --> 00:06:30,980 and the use of AI in document drafting, compliance workflows, and decision support systems. 80 00:06:31,200 --> 00:06:34,579 And we will always do so starting from a simple idea. 81 00:06:35,279 --> 00:06:38,579 Innovation is only useful when it remains governable. 82 00:06:38,579 --> 00:06:43,839 If these topics are of interest to you, you can also follow my in-depth analysis 83 00:06:43,839 --> 00:06:52,299 on my blog www.nchfab.eu and subscribe to the weekly NCHFAB newsletter, every Tuesday. 84 00:06:52,540 --> 00:06:56,559 Also available on my blog, where every week I share analysis, 85 00:06:57,160 --> 00:07:01,579 references and insights on privacy, data protection, artificial intelligence, 86 00:07:02,619 --> 00:07:05,140 cybersecurity and regulatory innovation. 87 00:07:05,140 --> 00:07:10,820 I look forward to joining you again in the next episode dedicated to legal prompting. 88 00:07:10,959 --> 00:07:15,019 Thank you for listening to this first episode of the NCHFAB podcast.