How Compliance Actually Scales with Hamza Siddiqui, Director of Compliance at Chime

Hamza, it's great to have you on the podcast. >> So delighted to be here. >> Yeah. So you once said it's not artificial intelligence, it's augmented intelligence. And you also lost 82 bitcoins from Pakistan to DC, two IPOs, and now running compliance at one of the biggest neo banks in the world.

Um you've seen the evolution of risk, regulation, and technology from every sides. So today, I'd love to unpack what you've learned about AI, fairness, and how to make compliance actually scale. So let's start from the beginning. Uh you were born in Pakistan, moved to the US in the 2000s, which as you mentioned was a very different America a year later. How did that shape the way you think of trust, risk, and systems today?

>> Yeah, you know, when when I first moved to the country, I was 8 years old coming the land land of things I've seen on TV. um getting to know everybody and then a year later 9/11 happens and very quickly the the reality that you were coming in with this childhood innocence changed very quickly and then to grow up in you know the post 911 era um and and to experience the world through that lens where trust became a pretty hot commodity and something you had to truly truly earn. It wasn't given anymore. uh shaped I think a lot of of my thinking and I think we see that in the world today you know the are we are we a high trust society are we a low trust society um and you kind of see that 9/11 happened the financial crisis happened you you saw digital money but I think the underlying context around crypto and the digital money is digital trust um and and that was that was like an interesting journey for me for me to think back to if I was 8 years old what I would what I would be doing in America. I thought I'd be a sports star or a celebrity and now I'm just a hunky dory, you know, compliance guy working in digital money.

>> That's great. So, that's super interesting to understand like how that initial setting really changed like your view on that. Um, but you've also done two IPOs and what actually changes for a CCO going from preIPO to public and what becomes a non-negotiable overnight? Yeah, it's it's really interesting to kind of see the evolution of a compliance program where you're in a very hyperrowth stage and and as you continue to go through your growth phases, as you continue to do big rounds, you know, the target on your back becomes bigger until you get to kind of the IPO stage where you're breathing a sigh of relief as everybody does. Oh, we finally made it.

And then you realize, uhoh, we finally have made it. uh the target is real tangible whether you're thinking about it from uh regulators, folks who are very happy to be latigious and and so your compliance mindset becomes less indexed on how can we grow, how can we support growth so quickly to how can we do so at a pace that is sustainable and how do we think about our risk appetite and in a way that was different um when you're in the private markets because your exposure is so limited. Uh but yeah, once you make it, you've made it. >> And uh and it's for all the good and bad, you're you're thinking through it very differently. It's it's less of a it's less of a let's build build.

Let's grow grow. It's uh how do we how do we defend what we have? How do we protect what we're doing? And how do we continue to sustainably think about this? >> Yeah.

And about thinking differently, there's something really interesting that when we chat last time, you were at Upstart. And so you joined there right before the IPO, helped navigate a CFPP, second no action letter, and launched their first non-personal loan product. And you said to me that you were a translator between machine learning engineers and regulators. What does that translation actually sound like in practice? >> Yeah.

And I when I think about the best compliance people, whether it's very technical people or very innovative products that they're working on, the best compliance people that I look up to that I think are the best practitioners that you see on all the talk circuits, they're the ones that are able to take these very esoteric or antiquated or very confusing laws and regulations and put them into the context for an engineer or a PM. Conversely, when you're kind of working in a space where you are helping develop innovative products, you are taking the very brilliant technical minds or the innovative ideas that are coming through and helping put them in frameworks that the people who are going to ultimately shape your destiny can understand before October of was it 2022 uh when suddenly everyone decided they were an AI influencer and you know top leading LinkedIn voice on on NAI chat with me. Uh we were doing some really sophisticated work at Upstart where we were taking a very novel concept. How do you move away from the FICO and the hard credit rules and give access to people based on their own merit besides just a small box that was created by somebody and that's been part of the industry for centuries. And we had to find the way to fit a very novel concept machine learning models and credit underwriting and help translate it to bank partners to our own internal people and then of course to the CFPB regulators to be able to say you know this isn't some kind of black box.

It's actually no different than the human brain. Um we we can help you understand how our machine learning models will be making decisions to determine should this person have access to credit uh based on our risk tolerance or not. How should we price things? And and it was the ability to work with the machine learning team who bless their hearts, right? They they they were willing to dumb down their very technical concepts to me and I was able to ask all the questions in this very safe space and then take that distill it in my based on my experience to be able to say you know here's all the features here's how they map ultimately to what sounds like pretty benign credit reasoning and why we would approve somebody or why we would reject somebody.

Um I think that was really applying that human element. I love my technical people but sometimes the altitude with which they try to explain their ideas doesn't necessarily translate for nontechnical people. >> Yeah. And you know like we we do AI and we talk to a lot of regulators and >> it is complicated sometimes to truly make them understand what we're doing and what's going on. >> And you were doing that as you said pre2022 like a time where AI was even more mystical to some people.

Um, so how do you actually do it? Like how do you make that black box legible for regulators without scaring them away cuz you were able to implement uh machine learning models at that time which I'm not sure every company managed to do and managed to convince regulators there. So how do you do it and how do you think it differs from today with LLMs and maybe more blackboxy models? Yeah, >> you know, it's there's a couple of different ways you can build products and and we see that across the industry. You can have a very buttoned up approach where you would think product is leading what ultimately comes to market, but it's somebody in risk or compliance and you have a very watered down version of what could have been a great product.

You also have on the other end a product that once you experience it, you're like, "Wow, this is probably for the average consumer going to be harmful." And I I think the best models that I've seen that I've had the ability to work with great leaders and implement on my own are where you're embedding compliance. It's part of the design process. It's part of the development process. And that means for when we were thinking about how do you take a very novel concept? How do you take something that's so technically difficult and ultimately bring it to market?

You embed compliance from day one. You let your technical people do what they have to do. you know, you want to design a machine learning model and you want to develop a new AI, go for it. Here is the sandbox that I think you're going to have to eventually come to play in and laying those things out for your for your development team and then letting them be creative. Let let the folks do what they do best and when they're ready to come to you, sit down, ex start teasing out what will ultimately matter to your regulators or frankly to the market.

Uh and for us this meant have whatever features you want. Have 500, a thousand. I the some of the teams I work with today and and some of the folks I advise it's don't constrain yourself. Do what you want to do, but keep the end goal in mind where you want to have a flying sophisticated product that's also going to be explainable, that's also going to be something that needs to adhere to these these things. And so the way to think about it is explanability is going to be at the core of any AI tool or an an ML model.

You need to be able to explain what's happening. How can I trust you? You know, how can I trust you to give me give me an unbiased credit outcome? How can I trust a AI agent to properly execute my BSA ML program? >> If I don't know what's happening behind it, I cannot account for kind of that unknown risk.

>> Yeah. No, 100%. And so the thing is from step now chime you're dealing with consumerf facing AI and so although you might have very you know understandable decisions very auditable decisions and understand the reasoning through them I would imagine that if you do one or two mistakes it's a very riskaverse space and um you might not be able to come back from there. Has there been any issues or has there been any blunders like that that you've seen either at these companies or others that you've seen was not um savable? >> I think of the issues as a spectrum really.

It's you know are you going to bring something to market that's going to be so terrible, so overengineered from a safety perspective that the user experience is going to be awful and it's going to polarize people. I think about when Google very early last year released a Gemini update and there were people trying to do certain prompts and they kept experiencing a a very strange outcome that you wouldn't expect >> from what should have been an otherwise easy use innovative product and that was I think a great example where you're overindexing on what you need to do to bring something to market. That was in my opinion a pretty big failure. Um, but on the other end, you you kind of had issues uh with a a uh was it Goldman and Apple card where they were unable to explain why they had gender discrepancies in their pricing and approving when they first brought the card to market. And and so I think from that perspective, you know, you lose a lot of trust.

I talked about that at the start. You know, we've h we've entered into this digital world where it's really trust that's digitized. And how do you approach that? And I think the way to do that as consumerf facing AI is when you're ready to launch, you're you've done thorough testing and you feel really comfortable that the worst actor isn't going to pick out something that you're doing that's incorrect. Uh from an ML model perspective, you know, you want to make sure it's going to be nonbiased.

From a consumer uh facing, you know, say personal finance tool, you definitely don't want to be encountering a scenario where you're giving me the worst advice. take out as much credit as you want. You know, go into debt if you have to. Uh, you know, the making sure the under underlying training data and ultimately what the output is is going to be something that's palatable. >> Yeah.

But, you know, when you're a startup, it's normal for you to try something you know you're going to fail hard and then go back at it. But in compliance, it's not something you can't do. So as a startup whether it's a fintech that wants to implement compliance or a rect startup how do you think you can you know have this balance between innovation and speed and yet safety and trust. >> Yeah. Yeah.

There was a great article the founder of Upstart wrote speed as a habit where you always are trying to move at a very fast pace. But to move very fast, you sometimes have to consider other things in mind. It's no different than when you're say going out for a run. You could sprint your whole marathon and you're probably not going to finish. Pull your hamstring.

ruin yourself in some horrible idea because you were trying to move too quickly or you could try to walk your marathon, probably get picked up by by the last cart and kind of not finish. Um, and and so as a startup, even as a public company, the way you want to think about speed and the way you want to think about innovation, I think has to come down to how fast can we go at a pace that's going to be comfortable. Should we be shipping LLM models or should we be using untested ML models and credit decisioning? Probably not. Should we wait to train our AI for over the course of a year and eventually bring a product to market that 20 other FinTechs probably already have?

Probably not the right approach either. So, I think it's finding that right balance of have we done enough testing to have a level of confidence this isn't going to break. This isn't going to harm our members or or or our consumers. And is it going to detrimentally impact our our image? >> Yeah.

And in something that is so important in FinTech as compliance, but that isn't necessarily core to the product as in the product itself. What is the your thoughts on build versus buy? >> Yeah. Yeah. That's a challenge I think every leader faces whether you're in compliance, an engineering org.

It's more often than not you're thinking I can build it and that's just that's just not going to be scalable and it's not going to be sustainable >> there. When I put on my compliance hat only because I'm not a technical uh leader, I think about all of the obligations we have in this space, whether it's BSA, AML, whether there's other consumer protection requirements. And the only way to execute at scale is through leveraging technology, is through leveraging tools. And then when you're facing that decision of should I build or should I buy, you kind of have to rely on on the fact that you're trying to support innovation. And in in the companies that I've been at, speaking about speed, speaking about velocity, you can't take 15 to 18 months to build a really smart, sophisticated tool to support the growth that has to happen in month 0 to two.

And that's kind of where you start looking out into the marketplace. And then when you're looking to buy, you want to look at and understand are these organizations bringing to me the scalable trustworthy solution that's going to help me meet my bank partner requirements, my own internal risk appetite. Is it going to release something? I like to joke about, you know, my nephew is the end user of a lot of the products I've been able to build in other family members. Is it going to harm my nephew or is it going to harm my mom?

Um and and and that's that's kind of like the decisioning that really goes into it. Cost factor is always a factor, but it's just what has been built. Can I trust the product? And that's where it becomes really incumbent upon on the person who's building things to to kind of show me, hey, we've built a great solution and you can actually trust us. There's been an article, I don't know if it was BCG or McKenzie, that said like 95 98% of enterprise AI pilots don't go through.

Why do you think that is? And I don't know if you've had such pilots at Chime. Um, and can talk about, you know, why you really think that is? >> Yeah, I I did read that article. I thought it was hogwash.

>> I was like, I'd love to understand your methodology more. There has to be some disclaimers here. Um I I think there's a lot of exciting things that are happening in just generally enterprise AI space. I if you asked me at the end of two quarters ago what tools I think would be able to help me and if you asked me to survey the landscape again and say all right here's here's is that tool that you selected still something you'd want to use today and the answer is going to be no because something new and better has come to the marketplace so I think pilots are a great opportunity for for enterprise company and and companies to really figure out is this going to fit my use case Is this good enough or is this also futurep proof? Pilot pilots are a great way to experience that.

Maybe that's contributing to that high failure rate, but I but I wouldn't necessarily classified as a failure as much as discovery. You know, I may I may test five tools, seven tools and say, you know, I actually really like one. That didn't mean the other five or six were were discarded because they were AI and I didn't find use for them. It's just I found the one solution that worked for me. um can't I'm not going to speak to specifically maybe what we do or don't do at Chime, but I will say in in a lot of the companies that I've been working with and advising, there is a real energy and a desire to find tools to augment the workflows, right?

You how do you bring on something that will make me a 10x compliance person or make me a 10x engineer? >> We'll need to chat after the podcast. But something I'm super curious about is and we had a chat about this before. Um we've worked with we work with a lot of companies and there's an immensely fragmented environment in which compliance teams work in. Um we talked about the fact that unlike you know sock 2 hea compliance with companies like Vanta there's never been a runaway winner for BSA ML compliance.

Why do you think that is? And do you think now with the advent of AI this could change? >> Yeah. Yeah. Big big congrats to the team and the backers at Venta.

I mean what a incredible success story. If you could see that replicated anywhere else it'd be miracle. Um BSA AML I think is something that has never gone out of flavor. uh regardless of say administration change, you've always seen some of the biggest consent orders revolve around uh uh this regulatory uh hurdle and and I think when you think about how much exposure banks and fintex have they want to be really careful and and that means if I if I'm a old school compliance officer I'm going to trust my team to execute my BSAML program rather than outsourcing it to consultants. or even rect.

I don't know how it works. I'm not going to risk my credibility, my bank's credibility, and for what it's worth, take on kind of the personal fines that accompany these consent orders because I can't trust you. >> I think that landscape has started shifting a little bit. uh Rectex have a I think started doing a better job of demonstrating upfront how they are trustworthy vendors and and and that shift I think will enable uh a clear winner where you have you have more and more banks not just FinTech who are leaning into the rect solutions to be able to say we have a lot of SAR filings we have a lot of things we have to still meet the obligation for regardless of if you agree or disagree with kind of certain Finen Finson thresholds you you still have to do that job >> and I do not want to go into yearly budgeting planning and saying I need a 10x increase in my headcount to be able to support whatever merchant activities we're doing or the new correspondent banking activity that we're doing. I would love to be able to bring in a tool and I think having new rects that are leveraging AI in a sophisticated way that reduces the need for me to hire more people that come to me and say here's an auditable demonstration of what is actually happening with this human to tech replacement that is going to drive you to feel comfortable to say you know all of my SAR filings are going out as expected that all of my transaction monitoring is occurring in the way that I would expect this person with 5 years of experience uh and I can audit it if I need to.

>> Yeah, I think it's interesting because when you think about it like so to compliance is a checklist that you need to hit in order to get that certification. BSA AML is completely different uh beast. um every team might have different things that they have to go through for different types of onboardings, for different types of customers, for different types of geos they operate in. So it isn't as clear of a checklist to be like, hey, we're compliant and we know we're doing it in a great way. And I feel I don't know out of your experience that maybe you know legacy systems have been very static um rulebased region focused um because of how complicated BSA AML is and now with AI we can get solutions that are much more tailored to an organization.

Yeah. >> And so do you think like if your rect start today what would you do in order to win the market? What do you think is the top one to three things that um would allow Reickstarter to sell to Chime for instance? >> I think you're hitting it spot on. Speed, trust, and cost.

I know every cost is like the unspoken elephant in the room, but you know how how much money are you going to save me >> in the context of how much growth my company's looking for? But it is really that speed to implementation. Certain legacy systems will take months to onboard and then you have additional quarters ahead where you just have to get it to understand your business which is constantly shifting in the fintech or much more tech forward banks with AI tools and some of the rects that I'm starting to see uh like a certain Sphinx for example it's having the ability to onboard a potential new agent as I'm considering a new market to be able to say here are some of the things that Singapore cares about which I did not consider if I was a US-based fintech only. As I'm looking to expand, I want to be able to onboard quickly, and I want to have the assurance and the trust that the rect knows the market, knows the rules, and is going to execute in a way that's going to protect my backside. Really, you know, you got to be able to save my butt.

>> Yep. And so now there's the other thing that we often see um us being in red tech you being chief compliance officer there's always friction with teams as in h compliance they're pain they're slowing us down um but now now I feel like on the contrary compliance can become a growth enabler um what do you think about that >> yeah I think that's spot on when I think about how the characture of a compliance officer is and how I try to operate. It could not be more diametrically opposed, right? I think speed is a good thing. Truly truly believe speed is a good thing and to be able to support kind of the marketplace today and how quickly something say in the influencer world is changing and how your marketing strategy has to shift.

Am I going to sit there and review 50 videos that need to be able to go out in a week? No, absolutely not. There's other tools that can help augment that workflow. Am I going to want to launch a product and limit it to say only a thousand people because I don't quite know what my fraud rules are going to look like or how I can do transaction monitoring? That product is dead on arrival, man.

By the time it's to market and something truly innovative, >> within a week, somebody else is going to release something that's probably not going to have all the safety constraints I'm worried about uh and and eat all of my market. >> Y >> So, compliance has to enable that growth. I said 10x compliance person 10 SF loves the concept of a 10x engineer >> having the right compliance hire who's thinking hey how can we manage risk which is a spectrum it's never black and white >> and how do we enable that growth is going to be so key and to be able to do that you need tech tools to augment the work that you're doing I I want to be able to do all of the things a traditional bank would expect me to do but I only have 24 hours a day and I a cost center. Y >> um and I can't do that with headcount and I'm going to do that with tools and you need a compliance person who is forward thinking who is willing to think through the risk pro risk protocols risk processes risk appetite and say you want to go to market and you want to release a very interesting generative AI tool let's release it here's just highle concepts give me an agentic AI tool that's going to do XYZ texting and let's get it out there >> yeah No 100%. But the thing that I find interesting in compliance is one is the risk appetants um that obviously um differs based on the organizations but usually when you have a product you have the vendor and the client but here there's a third party there's a regulator and what we've seen is sometimes the systems you use in house is as much like the use you get from it but also the satisfaction that you get from regulators from them whether you know, nice activized.

They've been there for decades and regulators love the logo. How do you think this can actually can is this necessarily a break for rect startups? Will it slow them down or is there a way to convince both the end client like Chime and the regulators as well? >> Yeah, you know, that's a really good point. regulators I think for a very long time didn't see innovation I wasn't I wouldn't say as a good thing but certainly not with the same rose tinted glasses that a lot of us in the tech world thought >> I mean you saw this in bank charter applications I mean from the great recession to maybe earlier this year the number of bank applications or trust applications were almost non-existent and we've seen a deluge of them starting this year because the regulators have started signaling there is a favorability ility um towards new products, new banks, new approaches.

And I I think Regex, whether you're new or even if you're still in Actimize, this is an opportunity for you to notice that the regulators are signaling their understanding that the world has shifted from when these laws were first written. the world has shifted since maybe even the earliest iterations of the FFIC exam manuals and that they need to think about a new world and and they're welcoming this approach tepidly but they're still welcoming it and and it's I don't I don't blame the regulators right you know everything moves fast but they're thinking in the long term to really take advantage of I think this shift in tide whether you're an old rect or a SAS you know depending on how actimize wants to consider themselves. Um, you have to be willing to speak to the regulator and have them understand your technology. Don't use the latest buzzwords, man. I don't I don't care if it's generative AI.

If I put on my regulatory hat, I need to understand what it is, what it's doing, and how it's going to be compliant with with, you know, what I'm going to ultimately assess you on. Um, and and I think for startups that are trying to get into this very incumbent heavy space, forget the logo, forget forget the name. Say we're doing the exact same thing that your actimize is doing. Here's the tech and here's how it's actually the same or potentially better. And and it comes down to the boring philosophical things of here's your audit trail.

Here's the rules configuration. And I think when you start speaking that common language of the regulator, they're going to understand, okay, all right, all of the sexy buzzwords and all of the sexy marketing aside, it's all the same on the back end for me. Okay, good. Yes, your community bank can use a a smaller rag tech rather than needing to pay so much for a much more clunkier solution. >> Yeah, it's funny because I mean we are an AI startup.

We, you know, we are just over a year old. AI is core to our system. And I hate the word AI. I think that AI is just, you know, a tool. It's a way to build.

And every Starberry company is become going to become AI. It's as if people are saying, "Oh, we're like a code company." Obviously, if you're a software company, there's code in there, you know. >> So today, if you're a software company, there's most probably going to be AI. And last time we chatted, you said something I really liked. You said you prefer the term augmented intelligence over artificial.

So what does that mean to you and where does augmentation actually create measurable ROI? >> That's a I love I'm hoping if uh if in the next 3 to 6 months it becomes a thing where people are like yeah you know we're augmenting our workflows because I I would like to think it's because of the podcast and I've done it. I've made it as a cultural influencer. You know artificial intelligence is just a scary concept. It's been, you know, think about think about like the earliest movies around AI.

It's evil, scary, you can't control it. I we don't have that in the marketplace today. We don't have that, I think, potentially for the foreseeable future. And we didn't have that even 3 years ago. What we have had is, I think, a tremendous leap in technology that is truly able to augment the work that I'm doing or augment my way of living.

I love the word augmenting intelligence because I'm not create using an artificial tool or creating a new workflow where it's artificially dependent outside of my constraints. It's an extension of me. Yeah. >> Uh as a as a compliance officer, it's an extension of my brain and for my team, it's an extension of the work that they're trying to do. Um it augments and makes you faster.

When you think about the number of SAR filings, it's not as if it's artificially increasing the number of SAR filings. It's not as if it's missing things. It's just extending my workflow. >> Yeah. >> I think the way you measure uh I think the success of your augmented intelligence is how much more efficient is it making you and how much your output has increased.

>> Yeah. >> You know, one side of the equation has to remain pretty static. I want my team of 10 to remain say a team of 10 but I want their output increase from you know a hundred pieces of you know media reviewed from a marketing compliance perspective to 400 500 and the only way to be able to do that is to augment their work there's nothing artificial about it >> agree I really agree and so today compliance teams are overwhelmed by tens of thousands of false positives that they have to manually sift through and it's just you know the work in which you just click Yes. No. And uh it's it's um not the most gratifying work.

And so now that we're able to potentially really be an enabler, be um an augment there, how do you think this will translate to growth as in what are the teams going to be doing that will actually expand the business? >> Yeah. Yeah. I think too many compliance people, risk people, operations people, even people maybe in in kind of the product or or or technical spaces sometimes see their role as processors. It's been done this way.

I have to execute it this way to get this output. And part of that is just because of capacity constraints. You have to get so many things done. You're going to have to devote so much time to reviewing your code to be able to get it to commit. You have to do so much time reviewing your SARS so that you can meet your filing deadlines.

You have to do so much material review so you can get it out the door so you can support the latest influencer campaign. By by using augmented intelligence tools, um, your team is able to get a lot of that processing done faster. So they're not devoting, you know, 80 to 90% of their time just doing things. Now your team's going to have the chance to stop, think about why they're doing things. Is this enabling my team to think more strategically in how they're approaching a workflow?

Is our, you know, transaction monitoring threshold accurate? Should we really be firing off 300, you know, transaction monitoring alerts? Do I really need to flag this much, you know, language in in a in a script for an influencer campaign? Can I take a step back at my disclosure approach? You really can't do that if you're under the constraint of it's coming my door.

I have an SLA and I got to get it get it out the door. Now you your team can sit back and think, hey, how can we enable this growth? >> How can somebody whose uh life depends on shipping code take a step back and think I'm not just executing against whatever my PMs told me the requirements are. uh I want to be able to iterate and provide suggestions back and say, you know, I was building this, everything looks good, but I have a new suggestion. It kind of enables that creativity that you don't otherwise see in in in these roles that aren't meant to be creative because of the constraints.

>> Yeah. And I don't want to age you, but you've seen regulators evolve over two decades. How close are we to seeing AI specific compliance regulations in the US? I think we're starting to see that at the state level certainly um and they're more mo than AI based. I know there's like a little distinction but there's real tangible conversations happening now on the hill um in state chambers to say what is the impact to our consumers?

What is the impact to the people that live in my state and how do I protect them against AI? I think the conversation to date has been from a place of fear and I think this is where rects consumer AIs can lean in and help educate the ultimate decision makers to say we are excited about the promise of AI but we also understand some of the risks for the financial services space. It's how do we think about modernizing AOA? >> Yep. >> In a way that makes sense for AI and ML tools.

How do we think about the way money moves today and how our filing and transaction monitoring system has in some ways failed to keep up? And where can AI help? And where does the guardrail need to exist to say, you know, this augmented tool or this machine learning model can be leveraged provided it has these fundamental criterion. There's there tends to be this desire to be very bullish on the latest technology, which nine out of 10 times I'm right there with everybody else. I'm like, "Yeah, give me these smart glasses.

let me let me use the the the the GPT wrapper and and try to augment my workflow. But we also have to think about how it's perceived to I think the more non-technical folks and understand that if they don't have a an understanding of how you developed it, they're going to think probably from a place of worse assumptions than you intended and you have to help educate them. So, I think we're close. I think we're going to start to see some more legislation trickle in across different states over the next 18 to 24 months. I don't think that the US Congress is in a place to get any significant landmark legislation through for quite some time.

>> They did crypto. So, it's >> the Genius Act was a surprise, I think, a welcome one for for a lot of people in many ways. Um, but we're going to start seeing, I think, the industry uh react to how different regulators are going to set up frameworks. A couple of years ago, NIST came up with their AI framework. And I think that's probably how we're going to start seeing uh and get at least tea leaves read on what is ultimately going to make its way into legislation.

How are these standards going to be set and what will they translate into? And one thing I'm curious as AI gets better, starts making decisions. Um, so not just aing people, but actually taking on those decisions, >> who owns the risk, and is that going to be a big question for regulators, the state, and companies like you that need and try to onboard rects um and AI startups. if um if you're an investor, if you're a regulator, if you're a boss, you can't really hold a computer accountable. You have to be able to hold somebody accountable.

And and I I think the smartest leaders know that in whatever iteration of where they outsource workflow, low-level decision-m, the accountability lands with kind of the human or the person. And there's that concept of a human in the loop. >> And I think as we continue to see more sophisticated AI tools that are on that spectrum from just augmenting existing workflows to pushing eventually novel decision-m, you're still going to need, I think, that human element to sign off on whatever constraints are driving a decision, whatever constraints are leveraging the AI decision. um and have that accountability exist. From a regulatory point of view, I don't think I could be wrong.

I don't know how how the landscape will change, but I feel pretty confident in saying if you are a bank or a fintech and you say our chief compliance officer is actually this new AI tool, >> you are probably going to uh have a very tough life uh staying above the water. Um, and if you're if you're if you're a for what it's worth, if you're a company and you say, "We've actually replaced our CEO with uh with this AI tool that's going to drive all of our decisions going forward," your investors are probably not going to be happy because there is that lack of accountability. Yeah, accountability remains and for a very long time in the future, I think will remain the key human element. That's going to drive AI adoption, drive AI enablement, and drive AI accountability. >> Yeah, I think it's interesting.

Um, I always compare AI to self-driving cars >> cuz even though and especially for what reading compliance um even though we might make u nine out of 10 times better decisions than the humans and find like for instance we've been able at things to find more true positives that humans can find than whatnot. Any single mistake that we will make um will be much more dramatic. Yeah. >> Than if a human had done it. Same for like a driverless car.

It will have the accident rate much lower than human drivers, but when crews hit a person in SF, they basically went under. How do you think we're going to be able to change our perception of that so that, you know, we understand that, okay, AI does make potentially better decisions than outsourced teams and whatnot. and being able to excuse small errors being made here and there that are less common than with human teams. >> I love autonomous vehicles. I think they're fantastic.

>> As somebody who walks around the city, bikes around the city, I am grateful to see a Whimo than I am a human driver because of what you mentioned. It's not just at a lower rate. It is at a insanely safe rate that an AI uh AV vehicle will operate versus me behind the wheel. And I like to think I'm a careful driver. >> Yeah, >> I I do wonder um how much of this is is a speed question where we talked a little bit before about the the the it's a spectrum, right?

You you can either go to market with something totally totally unguarded or you go to the market with something so guarded it's going to be effectively useless. That Whimo could have been something that moves maybe five miles per hour and stops within 20 ft of anything. Totally useless. God-awful experience. Put me on the back of some guy's motorbike at that point.

I I I do think we as technologists and we as operators and people who are bringing these products to market have an obligation to the consumer, to the marketplace to say, "Here's what we're bringing forward. Here is why we believe it's going to be so much better." And I think continue to prove it out. I I don't think we'll ever get to a place where a whether it's a um you know self-driving car that hits somebody and has zero repercussions. I don't think that's going to happen. And I don't think we're again back to my whole artificial intelligence CCO or CEO.

This is going to happen. But I do think we need to do a better job of explaining to to our fellow consumer of why this is ultimately a better choice and a better uh experience than what we've had for the last 20 or 30 years. There's a lot of economic anxiety, a lot of the fear of the unknown. And I think when you just start pushing things on people, they're going to default to areas where they've been taught something through a movie or what they've heard uh in like the latest policy paper or have read a study from McKenzie or Bane that says 85% of AI's adopt adoption is is a failure. So don't even bother checking out those tools.

>> No, I love that. Listen, that was really interesting. I do have to zoom out and I do have to talk about this this story because doesn't fit in there but I I need to talk about it. Uh we talked about crypto about the Genius Act. You told me it la in our last conversation stable coins are useful.

Everything else in crypto isn't as much. And then you told me that story about losing 82 bitcoins. Um, and so I need to understand, I'll frame it as a question, but what did that experience teach you about, you know, UX custody and the real world bar for crypto adoption? >> Yeah, I have made my peace with losing 82 Bitcoin. Very very early days of of of crypto.

Um, I down I remember it was a fun little high schooler who had just discovered tour and I downloaded it just to see what is this dark web thing they talk about and I'm do exploring it and I remember going on a forum and somebody's talking about Bitcoin and this stuff and they're like, "Yeah, if you're interested in Bitcoin, I'll send you some. Give me, you know, just give me some like gift card or whatever you have." And I said, "All right, I got like a gift card that has like two bucks on here if you're like selling me a transaction." They sent it to me. Um, and I forgot about it. Totally totally didn't register. You know, you get a new laptop when you go to college.

You get another laptop when it dies 2 years later and you want to fit in with everybody cuz everybody on college campuses has a Mac and you had an old Windows laptop and you leave it at your mom's house. and then your hard drive corrupts because it hasn't been used and it's a Windows laptop and then you realize in was it 2015 2016 something called Bitcoin is back at I think it touched 20,000 something per bitcoin and you're like I know what this is I had it I remember exploring it early let me try to find this this wallet of about 82 bitcoins no luck hardress corrupted >> that's awful >> didn't write down any of the details in a in a reasonable place and you kind of, you know, make your peace with it until, you know, it hits 100K and then you're like, "Okay, I really have to try to find this thing." And I did. We we were recently helping uh clean out my childhood home and I found a piece of paper that had five very long, you know, mix of numbers and letters. And I thought, "What is this? What could this be?" It was actually just ultimately gibberish.

If it was some kind of, you know, 15, it was 18-year-old Hamza's cipher to like protect his data. I don't know what it was. >> So, it's lost twice forever. >> It's lost twice forever. I was uh I was that meme where you look out into the sunset sitting on on the swing.

That's me. What has it taught me? Uh, you know, you achieve a certain amount of zen. No. um you know it it it kind of highlighted the the pros and cons of of kind of what what what the underlying technology is.

I mean there's so much Bitcoin that's essentially lost forever. >> A lot of people talk about that as a feature. >> I think about it I'm a very privileged person. I'm pretty pretty okay with my life. >> But for some people, you know, Bitcoin in their part of the world is their only way to truly have financial freedom until it's lost somehow.

Somebody hacks it away. You lose access to it for whatever reason. It's gone. Um, it's it's no different. I know people love to espouse.

Bitcoin is better than cash. It's the same. Once it's gone, it's gone. You burn your dollars. The government isn't going to trust you and say, "Here's another hundred bucks uh in cash." You lose your Bitcoin, it's gone.

>> Yeah. No, 100%. And so, the last question about AI. Um, you mentioned your 11-year-old nephew, your 18-month old daughter, and that you're already think about how AI fits into your home. Where does AIA genuinely help you know a family today, your family today and where do you draw the line?

>> Yeah. Yeah. You know, that's uh that's kind of I was talking about it earlier. When I design products, I do it with the people in mind. When I'm advising some companies, I'm thinking about what are you building today that's going to make my daughter's life better, easier, safer, sustainable.

For for my nephew, it's what are you building that's going to educate him? I think there's a lot of opportunity for how AI can impact education and just the family life. I'm working with a few companies now that are exploring very interesting ideas, so I won't give all of them away, but I think having the ability to leverage AI to be essentially a home manager is going to be pretty exciting for me. My wife kicks butt in addition to all of the the things she does at home. I like to think I'm doing okay at my job in addition all the things I should be doing at home.

Uh and and we want to have a home that is as efficient as we try to make our workplace in some ways. And I think AI can be a great leverage in that. I would love to have AI kind of be more forward in the education that my nephew is receiving whether it's you know from traditional sciences to even financial education. AI is going to be such a great companion for that. I also think about the the fact that there is an opportunity for for AI to truly enable um a a childhood that looks very different than what the kids have today.

You know, how do you find AI that drives kids to be more physically healthy? I think there's the recent update from Chat GPT around some of the constraints they're introducing and removing. And that is an interesting approach of, you know, how do you introduce AI to kids in a way that's going to be safe? >> Cool. Hamza, a quick lightning round before we wrap up.

Um, most overhyped AI use case in fintech right now. >> Oh, most overhyped. Uh, kind of the personal finance assistant. >> All right. Most underrated compliance control that actually moves the needle.

Wolf. Uh, can I say me and my job? >> Perfect. I love it. >> Um, the hardest part of explaining AI to regulator in one sentence.

>> The hardest part about explaining AI to regulators is the translation of code into reality. How do you take the software into the physical world? >> One hill you'll die on in compliance. Compliance can be a source of good and I'll leave it there. >> Love it.

The one metric you actually trust to prove compliance efficiency. >> O um NPS scoremember or user satisfaction. >> The one thing startups get completely wrong about working with banks. assuming risk tolerance. You know, that's that's a question I ask everybody on an interview.

What could you be when you grow up? If I wasn't in if I wasn't in compliance today, I think our old part of me was probably headed off to law school and probably somewhere litigating. I think that's >> that's a stereotypical answer, but it's probably the truth. >> Love it. Is there anyone else who should come on our podcast now?

>> Yeah. Have uh have you met this guy Sam? >> I don't think so. >> Uh he's running a small company called OpenAI and I think he might have some novel ideas. >> You should give me his number.

I'll bring him on next time. >> Uh but listen, Hamza, it was a pleasure having you here. I really enjoyed the conversation and yeah, let's keep it going and thanks for being on the podcast. >> Thanks for having me. What a great time.

>> Cool. Cheers.