- 2 hours ago
Trust, security, and AI in banking - can it really work? Ryt Bank and Universiti Malaya are putting Malaysia on the global AI map. Foong Chee Mun and Professor Chan Chee Seng reveal how research and industry teamed up, and what this means for everyday banking and Malaysia’s AI future
Category
🗞
NewsTranscript
00:00Hello, you are tuning in to Awani Review with me, Cynthia Ng.
00:13Homegrown digital bank, Wright Bank and University Malaya recently presented their AI research
00:19at the Conference on Empirical Methods in Natural Language Processing or EMNLP 2025 in Suzhou, China.
00:29One of the field's most respected platforms, a very proud moment for Malaysia.
00:34So today we will be looking at what the team have achieved and why it matters for everyday banking.
00:41Joining me on the show today are Wright Bank Chief Product Officer Fong Chi Man and University
00:46Malaya's Professor Chan Chi Sing. Thank you so much for coming on the show.
00:50Thank you for having us.
00:51So I'm going to start with the conference, quite a mouthful, EMNLP.
00:55I understand that it's quite a tough platform to reach and even to present, even for seasoned researchers.
01:01So for viewers who may not follow the AI world closely, what does presenting that actually mean about the work?
01:09What does it say about the work that you are doing, Professor?
01:13Yeah, I think like what you mentioned, I think it's a great achievement, not just for the countries,
01:17but as well as the homegrown talent, because a paper that we presented, which is the core banking functions for the right AI,
01:26eventually has been done by part of the university's students.
01:29And it shows that how the local talents eventually can drive the whole Malaysia ecosystem,
01:36that we always think that Malaysia can't do it for this.
01:40And as you mentioned, EMNLP is one of the toughest conferences in the world that is looking into natural language processing.
01:49So this is the second time that the joint collaboration between YTL and University Malaya have managed to get into EMNLP.
01:59Last year, we have our Malay MMLU, which is the first Malay benchmark.
02:03We also managed to put in into EMNLP, and this year we move one step further from theoretical perspective.
02:11Now we're going to a practical world to show the world that how this can be done,
02:16and it's been done in the most heavily regulated industry.
02:19So it's a problem for the countries.
02:21Okay, so let's get to the practicality of it.
02:23How does language-centric AI actually change everyday banking for the average customers?
02:30Is it faster, safer, is it more intuitive?
02:32What is it?
02:33To me, it's a lot more natural in the first place because following the advances in generative AI,
02:43there was a breakthrough about three years ago where in order to make the computer to do something for you,
02:50it used to take computing language such as Python and Java.
02:53Now we can do it in human natural language such as Malay.
02:56So you speak to the app in some ways.
02:59Exactly, so instead of trying to find it in the midst of all the menus, you get to speak to it directly,
03:05and it is able to converse with you and be able to execute banking transactions for you.
03:10I think this is especially important for Malaysia because we speak very differently from the rest of the world.
03:18So we speak Bahasa Rojak, we mix languages together and trust our app and to a certain degree,
03:28the Russian language model that we develop as well, Ilmu, is able to understand this, what we call code switching environment.
03:34Okay, that sounds pretty impressive. I have yet to test it, but I will after this interview.
03:41Okay, so while it eases transactions, it makes life easier for people, I think the core question here is how can you make people trust AI,
03:49especially when it comes to money?
03:51Trust is hugely important for a bank, it's the fundamental for a bank, and trust, because we are the first licensed financial institution to deploy this kind of AI in consumer-facing,
04:08we take a lot of steps in order to make it safe.
04:12For example, Write AI was designed to be a multi-agent system.
04:18When you convey your intention in natural language to Write AI, the first thing that it goes through is this particular agent called Guardrail agent.
04:28So what Guardrail agent is doing is to make sure that what the intention that you are conveying is in fact what it is supporting.
04:36So when you say transfer money to certain friends, it will be able to support this kind of intention.
04:45But if you say who is PMX, it probably won't answer you correctly, it will probably refuse to answer you because that's not the intention that you want to do.
04:54And that includes malicious intention as well, that's where we got it from malicious intention, so dangerous stuff will not get into the system.
05:06And then only that it goes into like payment agents, AI agents, and at the output of it, we have another Guardrail agent that got it from doing anything that is dangerous.
05:19And beyond that, we also have a network of international security AI researchers that help us to test the system in such a way or essentially in the computer science balance,
05:35we call it red teaming the system in order to make it safe.
05:39Okay. All right. I'm going to take this opportunity to just understand the science behind it a bit more, Professor.
05:45I understand you have, I think, maybe four key agents working behind the scene or is it more than that? That's what I understand.
05:50I think it's more than that.
05:50More than that. Okay. Walk us through, how do they work together in plain language and how can people picture their experience?
05:58How does it become seamless and safe for users? Because Right Bank is claiming to be the first AI-powered bank.
06:07So how, why? Why Right Bank? Why not other banks? And what led to this research? How did you guys become the first?
06:16Other than the genius of Dr. Chang.
06:19And the team. Let's give credit to the team too.
06:21I think, first of all, is a lot of trust that has been placed on using AI in this heavily regulated industry.
06:29As you mentioned, I mean, we have more than four different agents to look at different areas of the banking industry.
06:37And that's why working with the domain experts, the people in the bank, is very important for us because we are from the science part.
06:44So we are not really aware of the regulations and the process in the bank.
06:50So that's why this kind of a public-private partnership is relatively very important.
06:56So how we do it with this whole mechanism is that, I mean, first of all, we know what the industry requires to do a transaction.
07:07So that is why we have a couple of agents doing different things. This is relatively important.
07:12As Chi said, we then ensure that because when I said I need to transfer money to certain people,
07:20that executions could not be hallucinated. This is what people are worried about.
07:26So that's why we design a deterministic system. It means that it will exactly follow the instructions of you.
07:33There's no imagination. There's no hallucination. It didn't exactly follow on that.
07:38So that's why in the system itself, we sort of have a patented schema to ensure that all the information has been checked.
07:46It must be verified. And that's why a lot of people worry that, will it transfer my money automatically?
07:51Say, no, there's too much science fiction movie on that. That's why we always put humans in the loop.
07:59So it means that whatever you make a transaction, you are the ones who are going to click the approve buttons before the money can be transacted.
08:07So that's why not just that we use an AI system, but eventually we also included the human intelligence at the end of the day.
08:16So that is a very important part of it.
08:18I'm going to go a step further just to help us visualize.
08:21So currently, let's say I'm using a traditional banking app. There is a few steps.
08:28So I want to send something as money transfer and or to purchase something that's usually OTP.
08:33Yeah. And then you have to verify.
08:35Yeah.
08:36Say, okay. And then it goes through.
08:37Yeah.
08:38How is it different from RightBank's AI-powered model?
08:41Yeah. I think like first you say, you need to go through a sort of a manual that you have A, B, C, D, E process.
08:47But with RightAI, you just need to speak through it. I mean, transfer money. That's it.
08:53And the way that you convey the message, transfer money, it doesn't need to be even in a proper English.
08:58It can be Baselroja, as she said. You can use short form. You can use English. It doesn't matter that much.
09:05So that is how it is so different from other people on these services. And that's, again, for the security part, it will be almost the same because this is a regulatory requirement.
09:17So it's not something fancy that we could design by ourselves that, okay, let's remove the OTP. Nope.
09:23Anything that below 250, for example, it means that you must have a confirmation by the humans.
09:29But anything more than 250, then it means that you need to have two authentication factors out of it. So that's why this is where we do it.
09:36Essentially, all these security measures that you do with your regular CIMB or Maybank are grandfathered into the same process that we are doing.
09:46So beyond the AI security that we did, which means that our processes actually have stronger security versus the traditional banking.
09:56How so?
09:57Simply because we have the AI part that is protecting the transactions as well.
10:04And then everything else from OTP to face recognition are also being grandfathered into our regular processes.
10:12Okay. And I imagine that it's faster even.
10:15It's definitely faster to do simply because it's more natural to do.
10:18Even if it's not faster to do for certain segmentation, for example, like my mom, it acts as a coach when he's performing certain transactions.
10:29Like, for example, you would say, pay Michael, and then you say, well, Michael, what's his phone number?
10:36Or what is he doing now? What is his account number?
10:38So for the user, it sounds more like speaking to a friend rather than, you know, like a menu-driven type of transactions.
10:48This is all very impressive. I'm going to go a step further and ask then, this is currently the use case, the customer experience.
10:54What could come next then with this breakthrough?
10:57Just give us a sense of what can we expect in the next, I don't know, a year even.
11:04So one of the main thing is our continuous personalization.
11:09So right now, it is like you are going to a bank, speaking to a bank teller, but you're speaking to a different bank teller every time.
11:20What we want to do is that when you go to a bank, you speak to the same bank teller that, remember, like, hey, every beginning of the month,
11:28you're paying certain amount of money to your wife. Do you want to do the same thing?
11:33So this kind of personalization is what we are continuously be striving for.
11:39All right. Okay. That's very exciting.
11:41You just look at your personal assistance, financial personal assistance. Yeah.
11:44That sounds great. Okay. I'm going to pivot a little to talk about this partnership.
11:49Right Bank and University of Malaya. Just a quick note as well, Chi Man is also the CEO of YTL Labs,
11:56which is the research arm of YTL Group that developed Right Bank's proprietary AI technology.
12:02Okay. So talking about this partnership, right? So industry, academia, they don't usually work in the same timeline.
12:10And also in terms of mindset. What allowed this partnership to work smoothly? Maybe not so smoothly, you can tell me.
12:23Especially when you're... Nothing smooth with both.
12:26Especially when you're operated in a very tightly regulated environment like banking. Who wants to go first?
12:32Yeah, exactly. I mean, this is a very unique partnership, to be honest. And when she asked me, he has a project, I mean, and then he wants to work with it.
12:43And I always ask, are you sure? Especially in the financial industry, I mean, like you say, it's heavily regulated.
12:49So I think we complement each other. I mean, in this case, and it's a very good partnership.
12:53Because from the university side, we have the science part of it. So we have done some research, but it's always lacking of use cases for us to really deploy on it.
13:04And that's where then the bank side, which is the industry side. So it gives us the opportunity to look into that step.
13:13So in this case, I mean, eventually we grow together. This is very important.
13:18And then that also shows that how our students, I mean, can be using the theory that we've been learning in the university to be applied in a practical real world.
13:29So with this partnership, I think we bridge a lot of gap. This is the first part.
13:34And second part is that you also gain more trust from each tribe of us to say that, like, we're able to meet a deadline.
13:42And then so that we understand more when they say deadlines, what does it mean by deadlines?
13:46And when they say you need to reach certain level of services, what does it really mean?
13:52So this has not been in the university life before, especially for students.
13:57I mean, it's always about assignments, summits and case closed.
14:01But now you need to have better responsibility on that. So that is where I think this is good.
14:06That's a very interesting take from the university or academy perspective.
14:10What about from your perspective working with University of Malaya, a higher education institution?
14:15I think Prof has been a bit humble because Prof has always been many years ahead of his time.
14:23So I find it's my job and in the industry is to bring his many years ahead of his time to the market so that everyone enjoy the kind of innovation that he does.
14:36And that's not just within the innovation. It also brings the talent that he trained to be able to not just contribute in terms of science and academy,
14:47but to be able to contribute towards the industry and essentially to the benefits of people as well.
14:53And I strongly believe these are the kind of innovation that has to be more in the future.
15:02Because if you look at countries like United States and England, they are very heavy collaboration between top research universities and private companies.
15:17And that's where majority of the innovation comes by. And I think that kind of innovation is what is driving most of the economic expansion in United States right now.
15:29And hopefully that we can follow the same.
15:32Just turning back to Professor, this is of course the second time working with YTI Labs.
15:38What do you hope to see to increase this industry university collaboration? Because this is rare, at least to the public.
15:49What are some of the barriers in your opinion and what would you like to say to industry players?
15:54I think one of the biggest barriers in the industry right now is the lack of trust of our local talents.
16:00So I mean, so through this partnership eventually we proved that Malaysian talents is there.
16:05And you can start with the local universities, not just that you always must be appraised of people that are coming from abroad.
16:12But it shows that Malaysian education is world class.
16:17So you could trust our local talent to build your own ecosystem.
16:21So I think this is where it's always lack of. It's not about money.
16:24It's about trust and appreciation of our local talents over here.
16:30So that is, I will say that currently the major hurdle of the countries.
16:35Yeah.
16:36Did it take a lot of convincing from your end to work with?
16:39It's actually not too much of convincing.
16:41Firstly, I know Prof for more than 10 years and I always have confidence in his talent.
16:45But just want to reiterate what Prof said is that our large language model in AI Labs, EOMU,
16:55is actually entirely developed by researchers, Malaysian researchers.
17:01And 100% of those researchers actually came from Prof's AI Lab.
17:05So this is how valuable these kind of talents are.
17:10All right. Well, that's fantastic.
17:12I do want to talk about regulation. Banking, of course, is tightly regulated.
17:16This is of course ahead of, I won't say ahead of time, but ahead of the industry in some ways.
17:23Were there moments in the regulatory process where it felt like the rules weren't ready for something like this?
17:30What a main thing that when we first building Right Bank is that we tried to take things differently.
17:41So what we wanted, what we did was that instead of having compliance to be like the parent to the kids in the people who,
17:52who in Bank called the first line people, we designed the principle of compliance from the get go.
18:01So which means that when the product is being built, it's already reasonably compliant.
18:10Obviously we have our compliance team to check with the product again, but most of the time it is very compliant.
18:17And we involve the regulators from the first day when we design, when, and we convey our intention to the regulator.
18:26This is exactly how we design stuff. And we invite the regulator into the design process.
18:32And this is how we have a reasonably seamless time of getting, you know, this kind of product off to the market.
18:42And did it take a lot of convincing?
18:45We work with them. I think, I think the whole point is that we are working very closely with regulators.
18:51All right. I want to talk about responsibilities and red lines, if you may, when it comes to AI,
18:56because when technology starts influencing decisions, for instance, credit, credit assessment, right?
19:01The responsibility comes, becomes very, very real because you could basically opt up people who wish the technology thinks that,
19:09hey, you're unworthy of credit. For instance, this one example, in your opinion, how do you decide what AI,
19:15what are the areas that AI should not be doing? What are some of the ethical risks?
19:20Do you think that we should be considering right now when it comes to using AI in finance and banking?
19:27I think first and foremost, I think decisioning using AI has been in the financial industry for a bit of time.
19:37And my opinion is in such a way that AI is in fact a lot more replicable.
19:47And in fact, if you have a human to make credit decision, when the human is hungry,
19:54potentially the credit decisioning might be less fair than what it should be.
19:59Whereas AI is generally do not have any emotion attached to it, right?
20:04So it is essentially trained on the data that you had before, right?
20:10And it is a much easier process to detect the biases on your previous data and fix your AI decisioning.
20:24And in fact, recently I've been working on AI credit worthiness determination for the last 10 years.
20:32And we find that it is essentially with algorithms, it can be a lot fairer because you can check it independently
20:43versus potentially a human mind, which can be a lot more fickerish.
20:49Okay.
20:50Yeah, I think that what she said, data governance is very important over here.
20:55So that is where, I mean, a lot of times in vital AI labs and the university as well,
21:00we strongly looking at the safety of the data because this is where the data will be used to train the model.
21:07They want it to be, I mean, they want to be biased to certain people and so on and so forth.
21:13So that's why the trail of data, the transparency is relatively important.
21:19So like she said, we're able to audit trail up to the level of data to know that
21:25is there any bias that have been inserted to the model.
21:28So, but on top of that, I also would like to bring in a regulations over here that lucky or not,
21:36I think this should be have certain regulatory policies coming hand in hand with the AI,
21:42but not to stop the innovation of the AI, but rather to somehow stop people to misuse AI for the bad purpose.
21:52I think that is where it should be.
21:54Okay.
21:55Okay.
21:56So I do want to talk about ecosystem.
22:00So right now this EMNLP recognition is of course a very good first step.
22:06How do we then turn the spotlight to having a stronger AI ecosystem in Malaysia?
22:13This, I think the consensus or general feeling that we feel is that,
22:20okay, Malaysians are users of AI, but we are not yet as much a producer of AI.
22:26So how do we turn that around?
22:28And how does this research and this technology partnership,
22:32how do you take it to the next level?
22:34Yeah.
22:35I would say that honestly, we have already churned it.
22:38When YTL decided to invest almost $2 billion,
22:42I mean, in the YTL data center in Kulai, Johok,
22:46because that will give us massive competition power to become the builders of the AI.
22:53So that is already been done in the very first place.
22:56And now with the establishment of this private-public partnership,
23:00it means that we can groom more local talents
23:03and we have shown successful story in the right bank applications
23:07because it's fully built by Malaysians.
23:09So I would say that the view is already running.
23:12Okay.
23:13But now we want to have more people to be on board
23:16from different universities, from different industries
23:18to ensure that and to show to the world that we could do it.
23:23So that's why I would say that the ball is already rolling
23:26and we hope to have more people on board to make it much bigger.
23:31So the ball is rolling, where are you going to kick it to?
23:34So I think from a YTL AI Lab standpoint,
23:40we are building the infrastructure so that other builders can build on top of us easily.
23:46So I think for Malaysia to be an AI advanced builder country,
23:54what we need is that we need infrastructure that the builders can use
24:01and be able to, for them to solve their very specific problem,
24:05for example, in healthcare, manufacturing, financial services,
24:10and be able to use the substrate.
24:13We kind of make the bricks for them to build houses on.
24:17And what kind of houses is really up to them to build.
24:21What we want to do is to make it a lot easier and cheaper
24:25for them to innovate on top of us.
24:27So just to get clarification, this is a proprietary tool,
24:32I mean technology from you both,
24:34looking to expand it to other areas besides banking.
24:37Well, essentially, if you look at what YTL AI Labs is doing,
24:41is that we make our own last language model,
24:44including automated speech recognition,
24:47including image recognition technology.
24:53And these are the infrastructure technology for innovators to build things on.
25:00And one of the examples of innovation that can be built on top of this kind of technology
25:05is essentially the right bank.
25:08So that is an example that we want to show that,
25:10hey, this is how you can build something on top of your move easily
25:14and be able to solve problems for the users.
25:17So we are inviting, and in fact, there are more than 20 companies right now
25:22that are building on top of what we are providing in AI Labs.
25:26Across various sectors.
25:27Across various sectors from entertainment to automotive to educational to healthcare.
25:35All right. Well, looking forward to talk about that later on.
25:38But first of all, thank you so much for coming to the show.
25:41It's great to have you.
25:43Professor Chan and Chima, thank you so much.
25:44And best of luck with the work that you're doing.
25:47Thank you. Pleasure.
Be the first to comment