AI Governance Is an Illusion Without This

[00:01–00:03] Hello, everyone. Sarkas here with

[00:03–00:06] Verdict, and with me today in

[00:06–00:09] GRC Talks, Fabrizio Degni. He's

[00:09–00:12] an AI governance and ethics expert,

[00:12–00:14] and he leads AI and digital

[00:14–00:17] transformation in Italy's national agency

[00:17–00:19] for AI, and he also serves as the

[00:19–00:22] president of the Italy Chamber of Global

[00:22–00:25] Council and responsible for AI. Also,

[00:25–00:27] he's a researcher at the Politecnno di

[00:27–00:29] Milano and focuses on ethical and

[00:29–00:31] regulation frameworks for emerging

[00:31–00:33] markets. So if there's anyone who is

[00:33–00:35] thinking about uh the ethical

[00:35–00:37] aspects of AI, it's

[00:38–00:40] Fabrizio. So Fabrizio, thanks so much for

[00:40–00:43] joining us. Thank you so much for this

[00:43–00:46] space. Brilliant.

[00:47–00:49] Fabrizio, so we connected

[00:50–00:52] Primarily because of a post that you've

[00:53–00:55] made on LinkedIn, talking

[00:55–00:57] about a

[00:58–01:01] framework that maps out the major

[01:01–01:04] risks and concerns with with generative

[01:04–01:07] AI. And that post was was

[01:07–01:09] quite popular. And I would love to

[01:10–01:12] hear from you, what sort of made you

[01:13–01:16] write and be vocal about this?Yeah, I

[01:16–01:16] believe

[01:19–01:21] uhFor those of us involved

[01:22–01:24] in the daily

[01:26–01:28] technology and

[01:29–01:32] frontier innovation, before

[01:32–01:35] to be hyped or enthusiastic

[01:35–01:37] for the progress we see

[01:38–01:40] on a daily basis, we should be

[01:41–01:44] scared of what we

[01:44–01:47] don't see, of what nobody speaks about.

[01:48–01:50] BecauseEverything has pro and

[01:50–01:53] cons. If from a

[01:53–01:56] side there are benefits,

[01:56–01:59] uh we should ask which

[01:59–02:01] one is the cost?Nothing in

[02:01–02:03] life is free,

[02:04–02:05] unfortunately.

[02:07–02:10] For the cost we are paying

[02:10–02:13] to innovate on the AI front,

[02:13–02:15] I believe we should question

[02:16–02:19] each one if it is really worth it.

[02:19–02:21] So So what are some of the costs that

[02:21–02:23] you're seeing that that we're paying

[02:23–02:26] right now?I'm I'm reading a lot and I'm

[02:26–02:28] seeing a lot of predictions around what

[02:28–02:31] might happen. Uh What What do you think

[02:31–02:33] we're already seeing?There is a

[02:34–02:36] wonderful study called the 2027

[02:36–02:39] AI. Maybe we can

[02:39–02:41] later share with the

[02:42–02:45] people. Well, some

[02:45–02:48] researcher did some

[02:48–02:51] hypothesis for what could be

[02:51–02:54] this endless run to the

[02:54–02:57] AGI. AGI and

[02:58–03:01] ASI are the two steps for what the

[03:01–03:03] ANI, where we are currently now where

[03:04–03:06] we are supposed to have an intelligence

[03:07–03:09] with the same capabilities of the humans

[03:09–03:11] and with the superintelligence with the

[03:13–03:16] capability over humans. Now,

[03:17–03:19] just this comparison is pretty freaky

[03:19–03:21] because how you can compare the

[03:22–03:24] human mind, we still don't know how it

[03:24–03:27] works with something we are trying to

[03:28–03:29] compare with

[03:31–03:34] a different domain. Just this is strange.

[03:34–03:36] Okay, let's put on the same

[03:36–03:39] table humans live

[03:39–03:41] inside the context.

[03:42–03:45] what we are is what we

[03:45–03:47] feel. Every day we

[03:47–03:50] are different. Every

[03:50–03:53] moment we feel different because

[03:53–03:56] we are all of us. It's not just Brian.

[03:56–03:58] It's sense, it's

[04:00–04:02] memories, it's maybe

[04:03–04:05] emotion, it's empathy. It's

[04:06–04:08] millions of stuff that's happening

[04:09–04:12] in the same moment. And from the

[04:12–04:13] other side, we have an artificial

[04:13–04:16] intelligence trying to match us

[04:17–04:18] in the mechanical activities.

[04:19–04:21] Okay. But now the

[04:22–04:24] next step is to use

[04:25–04:28] artificial intelligence exactly as a pro.

[04:30–04:32] There are many applications already able

[04:32–04:35] to be like your psychologist.

[04:35–04:38] So ohh you see recently

[04:38–04:41] a big companyis

[04:42–04:45] planning to have inside the toys

[04:46–04:49] AI capabilities for skills. This is

[04:50–04:52] pretty dangerous because in my opinion we

[04:52–04:55] don't have any study on the long-term

[04:55–04:57] effect of this. When you

[04:57–05:00] look at the healthcare and

[05:00–05:03] there are some medicine

[05:03–05:06] to distribute before to go on the market,

[05:08–05:10] You wait years. You do some studies, you

[05:10–05:13] do research. Yeah, you wait years because

[05:13–05:15] it's not immediate. Maybe after

[05:15–05:18] two, three years something happens.

[05:20–05:22] But if you don't have this delay,

[05:22–05:24] decantation, time, you know, like the

[05:24–05:27] wine, if you don't have the time to

[05:27–05:29] understand what's happened,

[05:30–05:32] you simply get the side effects.

[05:33–05:35] And the side effects are

[05:36–05:37] unexpected. Now

[05:38–05:41] you understand that it's like

[05:41–05:44] every day we have like a gun

[05:45–05:48] because we are many companies on the

[05:48–05:50] same research

[05:51–05:53] fields and there is no a

[05:53–05:56] global policy or regulation

[05:56–05:59] or way we can

[05:59–06:02] say ethical way to approach on the

[06:02–06:05] research. hmm So

[06:05–06:07] I think it's a it's a really interesting

[06:07–06:09] point. I mean, you mentioned toy

[06:09–06:12] designers adding AI

[06:13–06:15] and and it sounds like a great product. I

[06:15–06:17] mean, I've got kids, it would be pretty

[06:17–06:19] cute to see my daughter playing with

[06:20–06:23] a doll that she can talk to and maybe

[06:23–06:25] could help her or teach her things. But

[06:25–06:28] obviously it would also collect a lot

[06:28–06:31] of data and information that that I

[06:31–06:33] probablyNot probably, I definitely don't

[06:33–06:36] want to be out there. So, you know,

[06:37–06:39] what what do you think, okay, let's say

[06:39–06:42] there's a designer, a toy company

[06:42–06:45] out there is probably working on it. What

[06:45–06:48] What do they actually need to do right

[06:48–06:50] now?Can they just release a toy like that

[06:50–06:52] that's maybe connected to Open AI, then

[06:52–06:53] everything that the child

[06:55–06:58] interacts goes online?Is there anything

[06:58–07:00] now preventing them or limits what they

[07:00–07:02] can do?Like today, I'm talking about you

[07:02–07:04] know recording this in July 25.

[07:05–07:07] Is there anything preventing them or

[07:07–07:09] limits them in any sort of way?My

[07:09–07:12] opinion, because this is a simple, my

[07:12–07:15] opinion, the time are not

[07:15–07:17] ready. There is a time for

[07:17–07:20] everything. This time is not the

[07:20–07:23] time, simple death. We cannot

[07:23–07:26] enforce something becausethe

[07:26–07:29] market or the

[07:30–07:32] the people want that because this is

[07:32–07:35] another problem, no?Why

[07:35–07:37] everything go?Because there is

[07:37–07:40] someone ask. And

[07:41–07:44] this problem is like a dependency.

[07:44–07:47] There is a phenomenon is

[07:47–07:50] called the FOMO, fear of missing out.

[07:51–07:54] You must be. If

[07:54–07:56] you are not there, you are not.

[07:57–07:59] And this is the same with now this chat,

[07:59–08:02] but everybody chat. If you don't chat,

[08:02–08:04] you are a loser, or you are

[08:05–08:06] something with

[08:07–08:10] less opportunity from your side than

[08:10–08:13] another, but this is not true at all.

[08:15–08:17] Even more so, I mean, even you see all

[08:17–08:20] the CEOs that are basically telling their

[08:20–08:22] you have to use AI or you're not going to

[08:22–08:24] be here becauseThey expect higher

[08:24–08:27] productivity from their from their staff.

[08:27–08:29] Correct. But yeah

[08:30–08:30] I

[08:33–08:35] like to say the AI is not

[08:36–08:39] the solution. It's like a tool,

[08:39–08:41] as many others we already have.

[08:42–08:45] If you have not a clear idea which one

[08:45–08:48] are the problem, which one is the method,

[08:48–08:51] which one is the process behind

[08:52–08:55] what you want to fix, you cannot simply

[08:55–08:58] expect, okay, I get the AI inside,

[08:59–09:01] it will be fixed at once.

[09:02–09:05] Not at all, because if you want

[09:05–09:08] to onboard the AI, your process,

[09:08–09:11] you must be aware you need first

[09:11–09:14] to have a process of assessment

[09:14–09:17] of your resources, just to see

[09:17–09:20] some technical aspects like

[09:20–09:23] oversharing of the files, people see only

[09:23–09:25] what they can see, nothing more.

[09:26–09:28] And the data, data

[09:29–09:31] are good quality because if the data are

[09:31–09:34] not in the good or you are not

[09:35–09:37] the way to collect in the digital way the

[09:37–09:39] data, AI can do very,

[09:40–09:43] very poor on that or can do a

[09:43–09:45] side effect, because

[09:46–09:48] companiesnow are

[09:49–09:52] trying to shift from a human approach to

[09:52–09:54] data-driven approach. Okay, Alura, we

[09:54–09:56] trust, you cannot trust people in data

[09:57–09:59] because at the end you must decide all

[09:59–10:01] the data on people. Okay, let's move to

[10:01–10:04] data. You focus all

[10:04–10:07] your business on data, but if you don't

[10:07–10:10] have a core strategy on the data,

[10:11–10:14] ohh it becomes a problem because

[10:14–10:17] you know exactly the garbage in, garbage

[10:17–10:20] out. If the data are not good,

[10:20–10:22] the output cannot be

[10:23–10:25] different. I want to

[10:27–10:29] just make sure, and I said, really

[10:29–10:32] because of your work with, maybe we can

[10:32–10:34] talk specifically about Italy, but I

[10:34–10:37] assume you you understand the

[10:37–10:39] European framework. Europe has been

[10:39–10:42] fairly fast in the past for

[10:42–10:44] adopting privacy. laws,

[10:45–10:47] GDPR, I think in 2018,

[10:47–10:50] and it was always fairly

[10:50–10:53] advanced. Even Even with fines and

[10:53–10:56] and enforcement, do you see the same

[10:57–11:00] approach being taken right now?I mean,

[11:00–11:02] if I'm if i'm an operator, if I'm a

[11:02–11:04] software company in Italy,

[11:05–11:08] are there any guidelines or limitations

[11:09–11:12] or laws or even policies that

[11:12–11:15] that have been implementedthat

[11:15–11:17] forces me maybe as a software company

[11:18–11:20] to act differently with my AI

[11:20–11:22] capabilities?I believe

[11:23–11:25] in Italy, but also in the other countries

[11:25–11:28] in Europe, this is enforced

[11:28–11:31] because we have authority and

[11:31–11:33] we have the famous European

[11:34–11:36] Regulation Act, where

[11:36–11:39] companies have some mandatory

[11:39–11:41] steps to follow up. So you must

[11:41–11:44] respect the rule. I am more impressed

[11:45–11:48] outside Europe, where there is not this

[11:48–11:51] regulation. Because the problem, and this

[11:51–11:53] is really my opinion, is not

[11:53–11:56] Europe has regulation, it is other

[11:56–11:59] countries has not. It's not

[11:59–12:01] the problem that companies are troubling

[12:01–12:04] to jump in Europe, because it's a market

[12:04–12:07] where you must be... No, No it's the

[12:07–12:09] opposite of the reason. All the countries

[12:09–12:12] must be ruled with rules

[12:13–12:16] And in this way you have also a way

[12:16–12:18] to have a competitive market. Otherwise

[12:18–12:21] the company Europe, what do they do?

[12:22–12:25] They should be limited

[12:25–12:28] because they follow the rule of the

[12:28–12:31] policies and the other ones not. We

[12:31–12:33] speak about ethics,

[12:33–12:36] principles and that one. It's clear

[12:36–12:39] ethicalwell and moral are

[12:39–12:42] related to countries and

[12:42–12:45] the cultural aspect. But there are some

[12:45–12:47] basic principle should be

[12:48–12:49] for everybody

[12:51–12:54] the same. We cannot really

[12:54–12:57] live in a country, in a world

[12:57–12:59] where everywhere

[12:59–13:02] someone has something different to say.

[13:03–13:06] Yeah. I think

[13:06–13:09] it's really interesting when we

[13:09–13:12] have, uh it's

[13:12–13:14] like there's not a new set of

[13:14–13:16] capabilities of what companies can do

[13:16–13:18] very easily, very fast.

[13:20–13:22] And you're right, it's not just about a

[13:22–13:25] company that is operating in Italy. It's

[13:25–13:27] actually, you know could be a company in

[13:27–13:30] Costa Rica, they launch a software

[13:30–13:33] and it has users in

[13:33–13:36] Italy and in Europethey have to comply

[13:36–13:38] with these rules. But it also

[13:39–13:39] that

[13:42–13:44] uh it's not going to be easy for Europe

[13:44–13:46] to to enforce rules on them. So then

[13:48–13:50] it's the ethical aspects of what guides

[13:50–13:51] the company owners

[13:53–13:55] to to operate ethically. That's the

[13:55–13:58] challenge. And I can tell my

[13:58–13:59] personal view, I think that, you know,

[13:59–14:00] we're in a capitalistic

[14:01–14:04] environment. And there's a race, like you

[14:04–14:07] said, there's a FOMO of how do we

[14:07–14:09] take this opportunity and grow as fast as

[14:09–14:11] we can right now and break

[14:12–14:14] rules and do what we need

[14:16–14:18] to be, you know, to be the best AI

[14:18–14:20] company, the fastest growing AI company.

[14:21–14:24] And I think it presents a real challenge.

[14:24–14:27] I want to take the conversation to one of

[14:27–14:28] the things that you mentioned, which is

[14:28–14:30] which is arounduh

[14:31–14:34] bias and discrimination. So, and how

[14:34–14:36] models are inherently

[14:37–14:40] designed to actually amplify the

[14:40–14:43] biases that we have. And

[14:43–14:45] I've seen it from my personal use when,

[14:45–14:48] you know, even if you want to generate an

[14:48–14:50] image of a woman, it's always going to be

[14:50–14:52] a very pretty woman. It's always going to

[14:52–14:53] be a woman that kind of meets the

[14:53–14:56] contracts of, you know, the the

[14:56–14:57] modernWestern

[14:59–15:01] standards of beauty. And And I think

[15:01–15:04] that's and I think that that probably

[15:04–15:06] within itself is is enforcing it. But

[15:07–15:09] what examples are you seeing in

[15:10–15:12] that AI is amplifying biases and and

[15:13–15:15] cultural discrimination?Yeah,

[15:16–15:17] I totally agree with

[15:19–15:21] your mind really. And this is another

[15:21–15:24] problem because model

[15:24–15:27] are nottrained with the

[15:27–15:30] specific dataset according to

[15:31–15:34] the engineer collected or the business

[15:34–15:36] required to use this specific

[15:36–15:39] dataset with the specific features

[15:39–15:41] and the specific weight. There are

[15:41–15:44] many parameters mostly we are not aware

[15:44–15:47] of. The famous black box

[15:49–15:49] uh effector.

[15:51–15:54] We have a ready dataset.

[15:54–15:57] collected and aligned according to

[15:57–16:00] some principle, then there is the

[16:00–16:02] part related to the training. And this

[16:02–16:05] is pretty interesting because the

[16:05–16:08] training is related to who trained the

[16:08–16:10] model. So if I have a

[16:12–16:15] some attitude for what I can

[16:15–16:17] be rigorous or

[16:18–16:21] strict on that, I have an opinion, I

[16:21–16:24] am human for that. When I

[16:25–16:28] interact with the chatbot, I'm not

[16:28–16:30] talking with someone agnostic,

[16:31–16:34] despite the topic. It's already

[16:34–16:37] with the position. There is a

[16:37–16:39] famous test did with the

[16:39–16:42] release of DeepSeek, when they try to ask

[16:42–16:44] to DeepSeek and the Western models uh

[16:45–16:48] same question. So same question,

[16:48–16:51] totally different perspective. But

[16:51–16:54] the problem is that people are

[16:54–16:56] not aware. This is a perspective.

[16:57–17:00] They think this is the answer. And

[17:00–17:02] it's very different. It's a point of

[17:02–17:05] view. It's actually kind of like humans,

[17:05–17:07] right?If you ask one expert uh

[17:08–17:10] what is the answer for a question, they

[17:10–17:12] will give you the answer that they feel.

[17:13–17:14] based on their entire life experience.

[17:14–17:16] You ask another expert, they're gonna

[17:16–17:17] give you, and they're always gonna say it

[17:17–17:19] with confidence as if this is the one

[17:19–17:22] true answer, right?So maybe in that--

[17:22–17:23] Yeah, but this one in computer is a

[17:23–17:26] computer, no?You are a human, this is a

[17:26–17:29] computer. Computer is no opinion, it's

[17:29–17:31] super parties, but it's not that. Umm

[17:33–17:36] So So maybe let's take it a bit more

[17:36–17:38] practically. Let's say that I'm the chief

[17:38–17:41] risk officer or chief information

[17:41–17:44] security officer. uh and I

[17:44–17:46] handle the governance in my in my

[17:46–17:48] company. What

[17:48–17:50] What do you think is my my biggest

[17:50–17:52] challenge now?What do you think is is

[17:52–17:55] really where where I would

[17:55–17:58] get stuck or I would need to put a focus

[17:58–18:00] on?Okay. There

[18:02–18:04] are many things you must be aware of and

[18:04–18:07] you should look at because at first,

[18:08–18:10] I believe you cannot handle

[18:11–18:13] everything. uh In the

[18:13–18:16] company you must have the filter with the

[18:16–18:18] request. And the first

[18:18–18:21] filter is why you need

[18:21–18:24] AI. Which one is the

[18:24–18:27] scope of your job?What do you

[18:27–18:30] want to do in this process?Because most

[18:30–18:33] of the time AI is not required.

[18:33–18:36] Or AI, which AI?

[18:36–18:38] Machine learning, deep learning,

[18:39–18:39] generative AI.

[18:42–18:44] level need different

[18:45–18:48] guardrails and effort and

[18:48–18:50] time. For that the first

[18:50–18:53] request and the first filter is

[18:53–18:56] before to put the hands. This

[18:56–18:59] is the last step. Okay,

[18:59–19:01] let's be clear of the business

[19:02–19:05] requirement. The functional,

[19:05–19:07] non-functional, domain requirement.

[19:07–19:10] All the things related to the process.

[19:11–19:13] They involve stakeholders, who is

[19:13–19:15] accountable, who is responsible.

[19:17–19:19] Must be something inside the process,

[19:19–19:22] cannot be simple. The request please

[19:22–19:25] check if the system is compliant, if

[19:25–19:27] there is something we must be

[19:27–19:30] careful about. Because, and you

[19:30–19:33] know that, AI is

[19:33–19:36] cross-domain. In the company, it

[19:36–19:39] is not only about one single stakeholder,

[19:39–19:42] it isan operation to be

[19:42–19:44] followed by HR, ICT

[19:44–19:47] compliance, um

[19:48–19:51] human resources, we see before, legal,

[19:51–19:54] of course, and the one

[19:54–19:56] analyzed the IT for the

[19:56–19:59] infrastructure. Because another problem

[19:59–20:01] is, okay, company come with the AI

[20:01–20:04] solution, but the AI solution is

[20:04–20:07] inside an infrastructure. Okay, ohh

[20:07–20:10] let's do firstthe assessment of the

[20:10–20:13] infrastructure you propose to support

[20:13–20:15] the business with your AI

[20:15–20:18] activity. Because AI is not

[20:19–20:22] run inside somewhere, and people

[20:22–20:24] now in your somewhere, because they think

[20:24–20:27] only to the last mile, the AI. That's

[20:27–20:30] actually, I think, an incredibly

[20:30–20:32] important uh point,

[20:32–20:35] becauseyou know, as a as a

[20:35–20:37] chief information security officer, a

[20:37–20:39] risk officer, I can really audit

[20:40–20:43] what I can, what I have access to, which

[20:43–20:44] like you said, is the last mile. It's

[20:44–20:47] whatever vendor maybe I'm buying from,

[20:48–20:49] but it's a black box. I don't really have

[20:49–20:51] visibility on the whole supply chain.

[20:52–20:55] Yeah, correct. Yeah.

[20:56–20:58] So it's also incredibly limited in what I

[20:58–21:01] can do. I would use a tool that is using

[21:01–21:03] AI,I can ask them where you

[21:03–21:06] store your, you know, your

[21:06–21:08] LLM, maybe do you run it on an

[21:08–21:10] independent closed server, but I don't

[21:10–21:12] really know. And I have no way of

[21:12–21:14] auditing this. No way of auditing. How

[21:14–21:16] would I ever know?All I see is the text

[21:16–21:19] output. It's not even with code

[21:19–21:22] where I can audit your code. Maybe if

[21:22–21:24] they let me, if I, maybe it's a very big

[21:24–21:26] enterprise deal, they would let you do

[21:26–21:29] that. But if I sign up to to a

[21:29–21:31] new tool with a free trial,

[21:32–21:34] uh and I give it access now to my Google

[21:34–21:37] Drive?That's it. Yeah, yeahI don't know

[21:37–21:39] where it's going. They send all your data

[21:39–21:42] and training the models. Yeah,

[21:44–21:46] it's serious. It's serious problem. And

[21:47–21:50] I believe we are not really

[21:50–21:52] taking care of the side effect of

[21:53–21:54] all of that. So

[21:56–21:58] firstly, it's super interesting. I think

[21:58–22:00] I think there's a lot of things to think

[22:00–22:02] about. here. In your

[22:02–22:05] work, what are you focusing on right now?

[22:06–22:09] A governance of the process

[22:09–22:12] with the colleagues and

[22:14–22:17] monitoring of the market on the

[22:17–22:19] trend and supporting the

[22:19–22:21] business and the project uh

[22:22–22:25] where someone asked about the AI

[22:26–22:28] andMost of the time

[22:29–22:32] we say, okay, maybe you can do

[22:32–22:34] without. Interesting. So

[22:35–22:36] based on what you're seeing right now, to

[22:37–22:38] wrap it up with the with the last

[22:38–22:40] question, I'm very curious to see,

[22:40–22:42] because you're you're seeing the

[22:42–22:44] regulation right now, you're seeing what

[22:44–22:46] companies are doing, and you're seeing

[22:46–22:49] what what steps are actually being taken

[22:50–22:52] to limit the risk. Where do you think

[22:52–22:54] it's going to be six to 12 months from

[22:54–22:56] now?Future?

[22:58–23:00] yeah Hard question.

[23:01–23:03] So according to Sem Altman,

[23:04–23:06] within July we should have

[23:06–23:08] GPT-5. So this

[23:08–23:11] GPT-5 should be something near

[23:11–23:12] the AGI.

[23:16–23:18] Global market and

[23:18–23:20] government on the same table.

[23:21–23:24] and think ethically and on

[23:24–23:27] the governance perspective, how they can

[23:27–23:29] speak on the same language,

[23:30–23:33] it will become an everyday

[23:33–23:35] competition. And the competition is

[23:35–23:38] something I believe don't pay anyone.

[23:39–23:41] Really, don't pay anyone. Fabricio Degni,

[23:41–23:43] thank you so much for your time and

[23:43–23:46] insights. It's been super interesting

[23:46–23:49] andI've read the the

[23:49–23:51] white paper that you posted and I

[23:52–23:53] strongly recommend anyone dealing with

[23:53–23:55] this, even if you're just curious to see

[23:56–23:58] what is done and what's really the risk,

[23:58–24:01] I think there's so many aspects that are

[24:02–24:04] so under discussed and

[24:05–24:08] should get a lot more attention and and

[24:09–24:11] it's coming up very, very fast. So we

[24:11–24:14] need to see what's happening and I think

[24:14–24:17] the solutionis not something

[24:17–24:18] that's going to come solely from

[24:19–24:21] governments and and countries. It's going

[24:21–24:23] to need to come from

[24:24–24:26] from companies themselves deciding to

[24:26–24:28] develop ethically, to build things

[24:28–24:31] responsibly, to be more transparent in

[24:31–24:32] how they do things, and also to be

[24:32–24:34] transparent in what they don't know, um

[24:35–24:38] because because their users need to be

[24:38–24:41] aware of whatof what they don't know,

[24:41–24:43] right?It's like, it's okay to say, it

[24:43–24:45] should be okay to say, actually there's a

[24:45–24:48] part of our system that works and we

[24:48–24:51] don't exactly know how. Uh, I know

[24:51–24:52] it's a very uncomfortable thing to say,

[24:52–24:55] but I think it's probably important to do

[24:55–24:57] that. Uh, Fabrizio, would that be

[24:58–25:01] okay if if our listeners contact you over

[25:01–25:03] LinkedIn?If we, if they have questions?Of

[25:06–25:08] course. Thank you so much. Brilliant.

[25:09–25:11] Have a great rest of your day, everyone.

‍