The Global Financial Regulator's Stance on AI

good afternoon I think these shows are that stability is a key uh key concept and uh we uh we had uh this morning our session of the uh Global Financial regul Summit four hours of uh open discussion uh and we touch upon many topics um usually our AI is considered by uh from the industry point of view how to increase competition in terms of Capital Market and but as a regulated Summit we took a slightly different angle we look at what's going on within the Regulators you know how they are can adapt AI what it means in terms of their own functions to what extent is increase their efficiency uh speededness and so on so we look at that and we look at it at a first step in a very uh concrete manner you know using looking at use case what who is doing what what happens instead of starting about high level speech very very concrete so I'm I'm I'm very pleased to have my fellow panelist maybe I'll I'll start with uh George from the uh Cypress commission chairman maybe you could tell us uh what you've seen in your own tradiction MH as use cases and then what kind of lesson we could uh we could draw from from from this first experience sure um um thank you thank you Philip for the question so uh first of all let me say that as an organization we understand the importance of the technology uh and we're embracing the technology um uh the markets have becoming very complex there there's a lot of technological development in the market so the use of Technology such as AI is imperative so we have been experimenting with the with the technology uh so one one experience one use of AI uh that we have applied at s it's a it's a it's a cloud-based Tool uh which is part by AI to allows us to integrate lots of volumes of data coming from the esma HS under specific European regulations so there's a lot of volumes of data that in the past we we did not have the capacity or the capability to analyze and now with the with the technology we're able to integrate all those data and not just integrate the technology would allows us to um do Predictive Analytics discover patterns anomalies in the data uh detections of any potential risks it incorporates uh NLP capabilities so there's a lot that we can achieve through this technology and this is part of a broader picture or a broader project I should say that we have which we call it the datadriven supervision where basically the data uh coming from all of these regulations which until now have not been analyzed to the extent that they should be analyzed will guide us in terms of our supervisory practices in terms of the risk categorization of entities and so on so this is one one use that uh has been very successful and positive for us until now yeah well thank you very much this is really useful how to uh use data and how to analyze data is a very efficient tool so maybe uh to uh luxur from the secur commission CL marks yes thank you and good afternoon uh to you all so before I actually go into the use cases I want to highlight two difficulties that we faced and I think we are not alone in facing those difficulties um before generative AI tools became widely available we tried to build our own Solutions with research projects with the university and this took an awful lot of time too much time for us um after generative AI tools were V widely available we were facing confidentiality and data protection issues that were also referred to by the previous uh panel actually and uh we have overcome those uh problems with a totally air gapped disconnected Sovereign Cloud environment on which the high performance latest Google AI tools can be used and so we can now use AI in a very safe uh manner uh having said this um we have two sorts of you use cases in our Authority the first series is to deal with very large legal documents ments um Luxembourg is one of the leading fund Industries in the world and we have 3,300 funds live with hundreds of pages of prospectus each and you can see uh that we should employ University graduates to review all this documentation so we can automate this secondly of course data uh analytics and data management reporting these days is very large and it's getting larger and larger and in order to be able to deal with the vast amount of data we use um AI um those are just two external use cases I could add to that um authorization process um not only reviewing large documents but also dealing with lots of fit and proper type of documents where this is also useful then we have internal use cases um ranging from minute taking uh in numerous meetings that we have as Regulators internal and external but also um you know dealing with um uh the the internal processes um we have embarked a few years ago on lean management uh implementation in the authority uh just like in a private uh company and we can make this lean management process much more efficient by plugging AI onto the lean management process so these are just a few examples thank you then uh the F ibim sure what kind of use case you can tell us sure thank you very much Philip um let me share with you a little bit um I I come from the financial regulatory Authority in Egypt uh who's the regulator for the non-breaking financial services but I'm be talking more about all the different use cases that we uh talked about and discussed from the different jurisdictions and it was actually quite intriguing to see that there is a variety of different use cases that covers the Spectrum all the way from uh the um generation of the jurisdictions uh and the regulations all the way or the legislations to the licensing process to the uh uh uh the uh uh digital reporting to the Anis of that all the way to the enforcement so uh whether it was a variety of different European jurisdictions as well as UK uh Asia there were really a lot of different use cases but the lessons learned to us or to me in particular was the fact that each jurisdiction would have its own um sense of uh uh figuring out what is the real value preposition based on their capabil ities and the competencies that they currently have and what they want to make up to so it's uh while there are a lot of different solutions that the AI and efficiencies that the AI can cover for the life cycle of what regulator can position the uh value on each regulator will be different and hence the importance of the continuous collaboration and we welcome this opportunity for us to exchange experiences and uh reach conclusions and shorten the the distance and the efforts in order to reach a cohesive framework for uh regulating AI okay thank you very much so then Emanuel javanis what about the uh FSR uh use cases I think we are kind of leading uh on authorization well we've got a number of Ed cases that we've been working on um firstly um I should mention the authorization process so we're using artificial intelligence to help us um analyze applications as they come in and when I say analyze applications so we receive a lot of different documents in relation to different uh aspects of an authorization process to meet threshold conditions which a regulator applies so for example Fitness and propriety financials structure um Pro processes policies systems and controls that you're going to implement so those come in a number of different formats so the idea is to get the AI to help our authorization officers to distill what the key risks are and key issues are within those documents and what may be missing in those documents and to be able to assist the authorization officer in being able to process the applications quite quickly uh another key aspect that we've been working on we call it the uh the reg buddy um we took our legislation a few years ago and we digitized that legislation so by digitizing it we then tried to um use an AI large language model to help us um in essence take a business model or business activity that someone proposes to do so say they want to do advising and arranging or managing assets or Collective Investments and put that into the system and then the idea is that the system will assist the firm in being able to analyze the full Suite of regulations and rules and guidance and be able to give say a compliance officer um a full distillation of what's required for them to focus on for their particular activities which we believe will help the community so that's more of a what we call a reg Tech solution rather than a subtech solution uh another key um piece of AI technology that we've been working on that we're fairly happy with is uh translation model so being able to use AI to very quickly help with regard to translation from Arabic for example to English so we're looking at AI from a number of different aspects we're also looking at it from a supervision perspective so from a compliance sence perspective but we do that because we have obviously some internal resources and skill sets but we also work with Academia for example we've been working with the mbz university for AI in relation to some of these tools so we don't just work in isolation obviously there are challenges um with regard to some of these things but uh we believe that technology generally but AI specifically is part of our regulatory toolkit going forward into the future without it we'll be left behind I mean industry is constantly moving forward and can see the benefits in Ai and so do we so we'll work in partnership with industry to not only understand it but also be able to supervise properly in the future oh thank you and this is very convincing in fact well we we we see that there are many use cases and um vast opportunities but I think that you mentioned challenges and of course well we cannot look at that with blind eyes there are risk behind that and the question next is to see how we can mitigate the risk in terms of governance in terms of data protection investor protection and so on there there risk so maybe we could start again the round of of of of of on this question uh how do you what what kind of risk you see and how you mitigate them briefly speaking yeah sure yeah uh so again some some of the risks uh that we encounter I think are similar to the risks that the industry is experiencing as well for example the explainability of the particular model that you're using uh any algorithmic biases that can be part of the model um the the model depends on the data quality so you know the data is the quality of the data is not there then there is an a model risk data privacy and security over Reliance on technology um so some of these These are types of risks that you encounter in the AI use for the industry but also for regulators and on top of that I should mention that uh there needs to also Regulators we need to develop the digital skills of the organizations um it's the the the the technology is moving in such a rapid pH and such a rapid Pace that um it's it's difficult to keep up to so there has to be ongoing training of our employees of our officers to be able to uh follow the new technologies and apply them in an efficient manner and we're doing that I mean we're constantly sending our people whether it's in um you know events locally internationally acmis in order to be educated on the new technologies and the second thing I should mention basically it's that you need as an organization to embrace this change the Dig you you you have to have a digital transformation mindset and it has to come from the top all the way down to all the people within the organization and that's not easy because there's always the resistance to any change basically right so and I should mention one one last thing that a lot of these things to implement them okay you might have a third- party provider but still you need to have the human capital inhouse to be able to apply the Technologies so they need time they need resources so a lot of times we're caught up in the day-to-day operations and we don't have that time to focus on these new technologies okay CL I'm I'm not going to add or to repeat what George just said they they the risks are the same I think key here is training of Staff uh I think uh you know they need to get up to speed up to a certain level in terms of prompt engineering how to ask the right question to the system um we also are facing obviously shortage of talent in computer science because that's not something typically a regulator is good at so we need to have appropriate talent in terms of data analysts data scientists and also you know people just knowledgeable in in the area and then we need to make responsible use of the tools for that we have a committee uh overseeing the use cases and also the use that is being made and we've also written a charter uh that all our staff have to uh uh obey uh because as a regulator of course we cannot afford uh to fall foul of any um rules or regulations or indeed any ethical standards TK anything to add sure um actually it's interesting know we're uh through a training process right now learning so I only have a couple of other thoughts maybe I can add to uh besides uh what was already mentioned uh from data quality uh privacy transparency of course accountability and responsibility of the use of the AI system is is a is a a key issue and maybe more in the Lesser mature jurisdictions the proliferation and the availability of the uh consumer and public AI systems uh there is a need for a regulation to say whether the financial service institutions and how they can use that as well as inside the regulator itself because you can't help but people try to be efficient because the technology is readily available uh so these two points uh besides the intellectual property uh ownership accountability this these are things that needs to be taken into consideration from risk perspective so to to address that uh we see a need to quickly come out with a regulatory framework that decides what is allowed in the um short term as uh and at the same time try to be flexible uh because things are going to be continuously evolving yes thank you you you use the word regulation so uh this morning we had a discussion on regulation not sure there was consensus of new regulation you know although it may be very high level like the EU act but for the financial sector itself for the sector itself more but best practice uh guidance and so so on so we are quite flexible but Emmanuel what is your view on the the how to how to mitigate the risk and uh whether we should uh take some action as regulators and well there's there's always uh this temptation to um over regulate or come out with extra regulations I don't think from an AI perspective at this stage we're in that position I think um one thing from today that U is really important is rather than worrying about even more regulations where we already have technology regulations uh we already have um guidance in relation to that we only just came out recently with a whole host of it risk management guidelines which touch upon these aspects already uh I think what's important and what came out of today uh is the cooperation and collaboration amongst Regulators globally not only on artificial intelligence but on all key risks and challenges that we face as Regulators so today we had Regulators from obviously the UAE the GCC but importantly also from Europe and Asia as well so we as Regulators need to work more closely together not only to um share experiences but also to share in how we manage risk and it's all about what are the risks in the utilization of AI uh but not losing sight as to the benefits at the same time so I think through this regulatory Cooper operation and collaboration whether it's bilateral or multilateral um through forums like today or through places like the bis Innovation Hub we continue to learn and we will apply that learning in practice through engagement with the industry but with each other so that's that's my main learning from today yes it reflect what we heard this morning about uh what the key word was uh collaboration International collaboration we have to work together share experience and then I would say progressive deployment learning by by doing somehow in a pragmatic manner so thank you very much thank you thank you thank you Philip