Sardine named a Leader in The Forrester Wave™: Financial Crime Management Solutions, Q3 2026

Learn More
FRAUDFORWARD
#95

エージェンティック・バンキングとAI主導による金融コントロールの未来

35分36秒

詐欺対策の皆さん、こんにちは。Fraud Forward へようこそ!

このエピソードでは、この議論が行き着くべき最も重要なポイントの一つに立ち返ります。つまり、きらびやかなAIの見出しを通り過ぎたその先で、金融サービスの現場における管理、ガバナンス、紛争対応、そして真の説明責任について語り始めると、何が起こるのかということです。

なぜなら、それこそがエージェント型バンキングをめぐる本質的な議論だからです。

銀行業務においてAIが実現可能かどうかという話だけではありません。

お客様がそれを実際に使うかどうかだけではありません。

しかし、銀行業務におけるAIエージェントが支払い、認証、紛争、そして顧客の信頼に影響を及ぼし始めたときに、それを銀行が責任を持って適切に管理する準備ができているかどうかが重要なのです。

この会話のこの部分で本当に良かったのは、単なるバズワードレベルの話にとどまらなかったことです。エージェント型の決済を、金融機関が安全に活用するために実際には何が必要なのかまで踏み込んで議論できました。銀行におけるAIガバナンスやAIによるトランザクション・モニタリング、銀行コンプライアンスとAIの関係、そしてAIを用いた銀行取引が顧客により近いところで行われるようになったときに、どのように信頼を守るのかについて話し合いました。

そして、それは重要なことです。

エージェント型バンキングが将来の金融サービスにおけるAIの一部になっていくのであれば、私たちはそれを明確な意図を持って構築しなければなりません。適切なコントロールを備えて構築しなければなりません。そして、銀行業における信頼のインフラが、技術に置き去りにされるのではなく、その進歩とともに発展していくようにしなければなりません。

このエピソードでお届けする内容

  • エージェンティック・バンキングが、従来のデジタルバンキングツールでは難しい形で、実際の顧客の摩擦をどのように解消できるのか
  • なぜ銀行業務におけるAIエージェントは、ガバナンス、説明責任、そしてコントロールに関する新たな課題を生み出すのか
  • AIによる支払い認証やOTPチェック、多層的なセーフガードが、AIを活用した銀行取引をどのようにより安全にできるか
  • なぜ銀行コンプライアンスとAIには、より強力なコンテキスト共有、優れた監査可能性、そしてより明確な責任の所在が必要なのか
  • AIによるトランザクション監視は、取引そのものの確認だけでなく、プロンプトや指示内容、エージェント主導の行動まで確認対象を広げる方向へ進化していく必要があるかもしれません
  • 顧客がすべての行動を自ら直接行うのではなく、AIエージェントに権限を与える場合、なぜ銀行の紛争解決はより複雑になるのか
  • 地域銀行や地方銀行のイノベーションが、主導権を手放すことなくエージェント型バンキングのメリットを享受する方法

このエピソードは次のような方におすすめです

  • 不正対策、コンプライアンス、決済、リスク管理、デジタルバンキング、または銀行業務に携わっている方
  • あなたは、エージェンティック・バンキングがガバナンスや顧客の信頼にどのような影響を与えるかを理解しようとしています
  • あなたは、チャットボット機能にとどまらない銀行業務でのAI活用を評価しています
  • あなたは銀行の不正防止、AIを活用した不正防止、そしてAIによるより安全な銀行サービスに関心があります
  • 実際の金融機関の中でAIを活用した銀行業務がどのように機能しているのかについて、より実践的な話が聞きたい
  • 銀行が統制に新たな抜け穴を生むことなく、どのように顧客体験を近代化できるのかを尋ねています

このエピソードを気に入っていただけたら、ぜひ購読して、iTunes、Spotify、YouTube、またはお使いのポッドキャストアプリでレビューをお願いします。

エピソードノートと主なポイント

メモの内容に入る前に一つだけお伝えしたいのですが、マーケティングチームから「ポッドキャストを見つけてもらいやすくするために、メモをある程度決まった形で構成したほうがいい」と言われました。これから書く内容は、その一部になります。

エージェンティック・バンキングは、実際の顧客課題を解決してこそ本物になります。

この対話で私がやりたかったことの一つは、エージェント型バンキングをもっと現実的で身近なものに感じられるようにすることでした。

理論上の話ではありません。

単に斬新に聞こえればいいというだけの未来志向ではない。

実用的。

なぜなら、銀行業務におけるAIの話は、そのあたりから多くの人がついていけなくなるからです。話があまりにも抽象的なままだからです。AI搭載の銀行サービスを、あたかも概念のように語りますが、人々が本当に知りたいことはもっとシンプルです。「それは実際に自分にとって何をしてくれるのか?」ということなのです。

だからこそ、子どもの世話を手伝ってくれるいとこへの支払いを例に挙げてお話ししました。支払いのタイミングは一定ではなく、つい忘れてしまうこともありますし、一般的な定期支払いのワークフローにはきれいに当てはまりません。まさにその点こそが、エージェント型バンキングが重要である理由です。より柔軟で適応的な顧客体験の余地を生み出し、あらゆるニーズを決められた商品機能に押し込めるのではなく、「意図」に応じて応答する仕組みを実現するからです。

それは、デジタルバンキングにおけるAIの在り方が大きく変わったということです。

ガバナンスは取引とともに動かなければならない

私たちが繰り返し議論したもう一つのテーマは、銀行業務におけるAIガバナンスは脇に追いやられていてはならない、ということです。

それは取引とともに動かなければなりません。

ここでティレンが指摘した点は、とても重要だと思います。銀行は、AIエージェントが銀行業務に入り始めたからといって、既存のガバナンス体制をすべて捨て去る必要はありません。中核となるバンキングプロバイダーは依然として重要です。既存の方針も依然として重要です。不正対策も依然として重要です。ただし今後は、なぜその取引が発生したのか、エージェントに何が指示されたのか、その行動が銀行のルールに合致していたのかについて、より良いコンテキストを持つことが、金融機関に求められます。

ここからが面白くなるところです。

なぜなら、エージェント型の支払いが本格的に拡大していくためには、銀行は単なる結果以上のものを必要とするからです。その結果に至るロジックを可視化できなければなりません。そうすることでこそ、AIによる銀行取引はよりガバナンスしやすくなり、より正当性を持ち、そして正直なところ、より信頼できるものになるのです。

信頼は誇大宣伝ではなく、コントロールから生まれる

これは私にとって最も強く印象に残ったポイントの一つでした。

信頼は、その技術が賢いと言葉で説明することから生まれるわけではありません。

信頼はコントロールから生まれます。

つまり、OTP(ワンタイムパスワード)による認証が必要だということです。つまり、多層的なスクリーニングが必要だということです。つまり、誰が何を開始したのか、どのセッションがユーザー主導で、どのセッションがエージェント主導だったのか、そして送金前にどのようなセーフガードが発動したのかを明確に記録しておくということです。また、銀行がいつ介入し、取引を停止したり、AIエージェントの行動を顧客の直接的なアクセスから切り離したりできるのかを把握しておくことも意味します。

AI を活用したより安全な銀行業務を実現したいのであれば、信頼をブランディングの一環のように扱うことはできません。信頼は実務として機能していなければならず、プロダクトそのものや認可レイヤー、そして意思決定の履歴に組み込まれている必要があります。

銀行コンプライアンスとAIには、より適切な文脈理解が必要です

リサは、この対話にとても重要な視点をもたらしてくれました。というのも、コンプライアンスチームがすでに理解している仕組みに、常に話を引き戻して整理してくれたからです。

それは重要です。

金融サービス分野のAIについて語るときに、人々が犯す最大の誤りの一つは、これをまったく新しい宇宙のように扱ってしまうことです。そうではありません。これは新しいインタラクションレイヤーであり、新しいデータ構造であり、新しいオペレーション上の課題です。しかし、根本的なニーズは変わりません。機関は、何が起きたのか、誰がそれを開始したのか、その取引が正当に承認され、コンプライアンスに適合しているかどうかを理解するのに十分な情報を必要としているのです。

だからこそ、コンテキストの共有がとても重要なのです。

銀行コンプライアンスとAIが連携して機能するためには、金融機関は記録を保存しておく必要があります。プロンプト、指示内容、セッションの種類、認可の履歴。これらがなければ、不正対策チーム、リスク管理チーム、紛争処理チームは、何も見えないまま意思決定を迫られることになります。

AIによる取引監視は今後さらに進化していきます

この議論を通じて、将来像に関する非常に重要な問いも浮かび上がりました。すなわち、「アクションの起点がプロンプトになったとき、AIによるトランザクションモニタリングは実際どのような姿になるのか?」ということです。

それによって審査プロセスが変わるからです。

怪しい取引を事後的に確認することと、その取引に至った自然言語でのやり取り、意図、エージェントの振る舞いを理解することとは、まったく別の話です。Tyllen は、エンジニアリングがすでに最終的なコードの出力だけでなく、プロンプトのロジックをレビューする方向へとシフトしつつあると語りました。同じ発想は、銀行の不正防止においても今後ますます重要になっていくかもしれません。

目標が変わるからではありません。

しかし、それは目標が変わるからではなく、証拠が変わるからです。

つまり、AIによる不正防止は今後ますます、「エージェントに何が求められていたのか」「その依頼をどのように解釈したのか」「最終的な行動が元々の意図と一致していたのか」を理解することに依存するようになるかもしれません。

紛争は簡単になる前に、まずはもっと複雑で難しくなっていくでしょう

私たちは銀行の紛争解決についても時間をかけて議論しましたが、正直なところ、ここはエージェント型バンキングが業界に対して、非常に短期間で現実的な対応を迫る領域の一つです。

なぜなら、顧客が「そんな承認はしていない」と言ったとしても、実際にはAIエージェントを承認していた場合、その次に何が起こるのでしょうか?

それは小さな疑問ではありません。

それは中核的な業務上の問いです。

リサは、認可が委任されたり、遅延したり、部分的に自動化されたりする場合に、紛争解決のあり方がどのように変わっていく必要があるのかを、とてもわかりやすく分解して説明してくれました。私にとって特に印象的だったのは、結局のところ答えは常に同じポイントに行き着くということです――それは「コンテキスト(文脈・状況)」です。顧客が何を意図していたのか、エージェントが何を実行したのか、どのようなコントロールが存在していたのかを、双方が十分に理解できるだけの情報を持っていなければ、紛争処理のプロセスを公正に解決することは、はるかに難しくなってしまいます。

だからこそ、ここでは標準化が非常に重要になるのです。効率性のためだけでなく、説明責任を果たすためでもあります。

地域の銀行は傍観していてはいけません

この議論の中で、コミュニティバンクや地域金融機関にぜひ誤解してほしくないのは、「エージェント型バンキングは大手だけのものだ」という考え方です。

そうではありません。

多くの点で、これは地域銀行や地方銀行がイノベーションを起こすうえで、最も大きなチャンスの一つになるかもしれません。ただし、各機関が適切な管理体制を整えたうえで取り組むことが前提です。本当の差別化要因になるのは、AIについて一番多く語ることではありません。最も安全で、最も賢く、最も信頼される体験をその周りに構築できるかどうかです。

そして、それは今でも人が中心のビジネスです。

それは今でも人間関係がすべてのビジネスです。

つまり、この議論は他のどの分野にも劣らず、コミュニティバンキングにこそ関係のあるものだということです。

AIが銀行業界にやって来る

このエピソードで特に印象に残ったのは、エージェント型バンキングとは、AIが銀行業界に導入されるかどうかを議論する話ではない、という点です。

すでに実現しています。

本当の問題は、銀行がそれを責任を持って構築するかどうかです。

彼らが信頼をシステムそのものに直接組み込もうとしているのかどうか。

ガバナンス、不正防止、認可、コンプライアンスを、リリース後の後始末ではなく、最初から中核的な設計要件として扱うつもりがあるかどうか。

なぜなら、不正行為はいまだに人間が主導しているからです。プレッシャーも依然として現実のものです。そしてお金が動くとき、特にAIを活用した銀行取引では、その結果は決して技術的な問題だけでは終わりません。信頼を揺るがし、安定性を損ない、現実の人々の生活を直撃します。

だから、これは決して金融機関が恐れるべきものだとは思いません。

しかし、これは彼らが真剣に受け止める必要があることだと思います。

エージェンティック・バンキングは、顧客にとってよりスマートで適応力の高い体験を生み出すことができます。不正防止対策を強化し、AIによる取引監視をより高度なものにし、金融機関がより安全なシステムを構築することを支援します。

Guests

ティレン
ティレン ビチャクチッチ
Paymanの共同創業者
Episode transcript
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
39:56
I think this has been great. I'm going to kind of jump back in here for a second, Lisa, if you don't mind. I really love the perspective, Tyllen, of having that where you build it for your bank. That it's not that you're going to come in with these, you know, black box rules of this is how it's done. It's really you own it. You build it to the specifications of what your customers are either asking for, what you're willing to give them access to at this time. And I think it's also a way to kind of safely onboard, right? It's like, we're not giving you keys to the kingdom, but we are going to test it. We are going to look to see, does this even work for our customers? Does this work for what solutions we have? If not, then maybe it doesn't make sense. But if we are looking to innovate and to bank our current customers, kids and grandkids, maybe this is the way to move forward. And I was thinking, as you guys were talking, I was like, you know what, I think there is another use case, at least for me. Right now, I have someone that, the external funds transfer process in my current online banking is a nightmare. So I have to go outside to a P2P app and I have to transfer money to my cousin who helps pick up my kids after school during the week. So every week I told her, said, look, I've got a lot going on and I'm gonna forget. Like no matter what, no matter the time, even if I set up a calendar reminder, if I'm in the middle of something work wise, I'm probably gonna forget. Please text or call me and don't ever act like you can't text or call me because you did a service and I need to pay you. And so I get these text messages on Saturdays. Hey, Haley, just a reminder, you forgot to pay me. And I'm like, my gosh. And so I can't set up a daily or every Friday a payment because She gets paid for which day she picks them up. Sometimes it's two days, sometimes it's three days. So I would like to have a notification from my banking app that says, hey, do you want to pay your cousin and how much? And ask me on Fridays. Ask me Friday nights at eight when I'm not working. And that's the one time for sure I can say, I know I'm not working. That would work perfectly. But yeah, so I just wanted to throw that out there. There's a way that agentic banking would work for Haley.
Tyllen Bicakcic
42:17
Yeah, I love that. mean, that's definitely possible right now. it's just, and to add on that point, think about what an OLB would now have to do. Pre-AI world. Now this becomes a product roadmap feature, right? Now we need to see where do we add this on the screen? Now we need to know how to like, you know, fight for resources and do all of this. This is where the shift in design, even with teams will come to where, okay, here's another intent. How do we support this intent? Do we need to connect another API? Do we need the AI to be able to make cron jobs? Very cool. If it does the ability to make cron jobs once there's not just that intent it opens up. It's like, okay, I need to remind me to check how much I should pay my vendors as an SMB. And you don't now need to build, a new SMB experience, all of this. This is where the bank can start to just lean in. into everything. One of our things that we work with our banks are it's like, how do we market this? How do we get people comfortable with it? Literally the marketing is just ask it. Just ask it. Like, you have a question. Do you think it can do that? Just ask it. It'll probably be able to do it or help you. And that's what we did with ChatGPT. Everyone uses it. Maybe the first couple of days, they realize the power, then they lag off a little bit. And then 15 days later, something clicks. and you just keep coming back to it and back to it and you start using it for more and more things. That would be such a beautiful experience for a bank to see for their customers.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
43:46
Totally agree. And I do think though that part of the other way that we were framing this originally, and I'm sorry, Lisa, if I kind of overtook this portion, but I just had another question to ask. But we talked about influence for the financial transaction. So I just wanna ask, especially as we're still talking kind of on that governance side, what governance structure should exist around these AI systems that influence the financial transactions?
Tyllen Bicakcic
44:16
One, as mentioned, all of your existing governance for financial transactions stays the same. You have your core banking provider, you have your policies and risk management there, that doesn't change. You have your online banking provider, that doesn't change. Now for the financial transactions, what you can provide from here is just more data. If it's initiated through an AI, did this, you We know the system requires an OTP check. Boom, there's determinism for you. The transaction cannot go through unless the user checked their email that was sent to them by your bank or through text message. To then plug that information in. Then from our side, this is where you have to build a very strong engine. It's not enough for us just to say, Hey, we're going to take your core banking provider. We're going to take their APIs. We're going to put it in an, in an MCP and we're just going to connect them to a plot model. That's not enough. have to build many architectures of AI agents to be able to make these decisions. To go and do this. And then I need to go back to the bank and I need to be able to show the compliance team. Like here's exactly how the agent thought about it from one. Here's the input that came in to was, this even allowed by the bank? We have our own input sanitizer that uses LLM, that uses RegEx, that uses sardine, everything to make sure this intent can even pass through. And then you see all these layers of decision-making and then you get that from the payment side and it's like, boom, here you go. You can make your decision of if this was good or not, let alone what you also get from your existing core. core provider, from your existing fraud provider, from your existing online banking provider. This is where you truly have to make that financial transaction feel safer, smarter. And one of the things I always say is we've just created the greatest technology ever.
Tyllen Bicakcic
46:09
With outside of fire with AI. It's like, don't be so scared. The technology is ready today. It's more than ready. It was probably ready last year, but today it's even better. Next year it's gonna be even better. And then how do you trust the vendor to be able to help you with these things and provide all these? This is where you truly need to understand banking also and not just be, you You have to have the banking experience. You need to understand what makes bankers feel comfortable because in turn that's going to make them feel comfortable with their customers.
lisa
Lisa Durnford
46:41
Tyllen, you just touched on so many things that... I think hit on different aspects of compliance and dispute management as well. know we want to talk about disputes. When I was first kind of thinking through these concepts of how we interact and how we manage the risks of agentic banking, it's helpful for me at least, and I'm sure some other compliance folks to tie it back to compliance systems we already understand and already use in day-to-day basis. How does this tie in? How does this fit in? all comes back really at the end of the data sharing and understanding that context. So you talked about, you know, how is that context shared? How is the influence or the intent, the instructions, you know, shared as part of that transaction? I think a lot of, you know, compliance minds would think back to the travel rule and how critical that is in terms of managing the risks and managing obligations associated with transactions, especially when those transactions do flag risky or kind of behavior that might require investigation, might require reporting, might require dispute resolution in the future. And without proper context sharing, we're really in the dark when it comes to addressing and solving for all of those things. So the travel rule as it applies to wires or really any payment, I think kind of applies here in the same sense that we are at risk if we don't have enough information. about that transaction, about the person initiating the transaction, the counterparty, all of the parties when it comes to monitoring, when it comes to risk flags, when it comes to reporting suspicious activity, when it comes to resolving disputes, it's really so key. And I think how you framed setting up that system to ensure that all of that data and context is shared is really going to help with that comfort level of feeling like I can manage these risks because I'm getting all the information that I need. That also touches on, I've heard this term many times of know your agent and we kind of talked about it a little bit before this recording, but I think ultimately it comes down to just understanding kind of who you're interacting with and where the instructions are coming from and how do I identify, you know, where the instructions came from and of course that could be solved for by ensuring that that proper kind of data package essentially is shared. with the transaction and identifying an agent led session versus a user led session and just understanding how that works. And I think it's, it is comforting, I suppose, for me at least to feel like, you know, there's layers to how we can implement this, you know, responsibly layers to how we can adapt our compliance processes that we're already familiar with, that we're already operating with. to just manage maybe new data sets, new data fields, and just new ways of interacting with our consumers really at the end of the day.
Tyllen Bicakcic
49:52
This is, there's something that you said that made me think about something that could actually help a lot on the fraud and compliance side, which is today in engineering, you basically assume the AI agent has done the code right. You know, so it, before what auditing looked like for engineers was, Hey, you go code and then let me as an engineer review your code and let me go line by line and make sure that it all looks good. As AI kept getting better and better, you kind of. absolve the need for that. And really what you ask for is, let me look at how you did your prompt. Let me see what you asked the AI agent to do so that I can understand, did you structurally ask it to do the same thing? And this is where you have, there was this new startup that just came out from the former GitHub CEO that started this to basically attach prompt logs to GitHub code reviews.
lisa
Lisa Durnford
50:45
Mm-hmm.
Tyllen Bicakcic
50:48
This is very similar to what you should start seeing in the transaction space, where it's like, assume the agent can definitely do the transaction. What was asked of the AI agent to go and do these things so that you can manage that shifts a lot of how you audit things. You're starting to audit natural language. You're starting to audit conversations. And then you want to see, does this fit the mold of what's a reliant, what's a request that should pass through? Yeah.
lisa
Lisa Durnford
50:52
Yep. (51:15) Yeah, exactly. And how does a tool like Sardine, how do banks risk tools adapt to monitoring instructions, prompts, natural language? It's a new same goal, but certainly a new way of monitoring for different structures of data, in a sense, to give you that proper context. Yeah, no, it's interesting.
Tyllen Bicakcic
51:38
This is why I would push start internally with it, because even with external, internally we have our engineers have cursor, they have cloud code. have our own open cloud AI agent instances running that could do engineering. If I was to just open this up externally, like I can't really audit even if they do it on their own cloud. and figure out what's going on. And if they do it on their own GitHub repo, that's the same problem you're going to run into if banks just start opening this up to everyone. Because that, that tech is still being built out and that's still ish. being developed on like attaching that identity, as you mentioned, like KYA, how do we attach a person's identity to a verified AI agent? Because at the end, who's prompting this agent to go and do these things? And how do we pass that information back? Because if I look at it right now, what OpenClaw opened up, no pun intended, is the ability for multiple people within an organization to act with the same instance of an AI agent.
lisa
Lisa Durnford
52:23
Yeah. Yeah.
Tyllen Bicakcic
52:41
For example, we have our ticket creation system for our engineers. It uses my API key. So it doesn't matter if I use it, if my co-founder uses it, if our head of product uses it. It always looks like the person that created that ticket was me. But that's not true because first, the AI agent created the ticket. And second, the AI agent might have been prompted by my head of product to create the ticket. So how do I audit to that level of identity all the way down? It's very hard externally. People are still building. We're building things to help that be easier, but it's going to take a little bit. But if you build it internally, you can always say, okay. This user is authenticated through my OLB. Boom. have their identity. They're interacting with my AI agent that I've supported. Boom. I have that. The AI agent is the one making the calls because Payment provided that information. So now I know Tyllen is the authenticated user because you should not be giving your bank account access to someone that's not authenticated. Right? So built-in trust, AI agents there, and then boom. That's that. I think those are the two levels of how to like think about this if you're a bank professional on where one's very scary and the other one is not so scary. very manageable and doesn't change much what you're doing.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
53:57
So you guys have teed up the next section perfectly, talking about authorization, right? Disputes and maybe even loss of ownership. So these are those operational realities. know, agentic banking forces the industry to rethink authorization. So the scenario could be a customer disputes a payment and says, well, I didn't authorize that. Yeah, but the customer did authorize the AI agent. The agent initiated the transaction. So then what happens? You know Lisa, I'll ask you first. How do you think disputes may evolve if AI agents are interacting with financial systems?
lisa
Lisa Durnford
54:35
Yeah, think the way that, I mean, a lot of what we've talked about is... kind of ties into this very well, but how we think about resolving the dispute and understanding the situation that we're reviewing is certainly going to change. Yeah, same goal in mind is, was this an authorized transaction and is this a fair dispute? And of course the merchant typically comes into this conversation as well. So I think with any, and I mean, we've talked about this in the context of PSD2, we're now three compliance. and ensuring strong authentication for different types of transactions, as soon as there's a delay between the instruction or the kind of thought being put out there by the consumer and then the actual transaction that creates sort this opportunity to feel like there's a gray area potentially between what was the intent of the consumer and did they get that and sort of was this transaction authorized by the consumer. And so there are frameworks under PSD2 to think about delayed authorization, merchant initiated transactions, know, having cards on file that you've pre-authorized. So maybe there's sort of an agents on file kind of concept, like you're talking about a bit, Thailand of having, you know, pre-authorized agents and payees in your system. So I think ultimately what I would say it comes down to is, you know, how do we adapt dispute resolution policies to close that gap, to ensure that we have enough context. When we are reviewing the transaction information, do we have enough context to close the gap between the prompt or the instructions that were initially provided by the person and what was actually executed by the agent? And if there wasn't maybe a human in the loop, and obviously that would certainly help to have that deterministic interaction of OTP, for example, if there isn't that, how would we get comfortable with feeling confident that the actual transaction matches you know, the goal or the intent of the initial instructions. Part of solving for that, of course, is having that prompt, having that context shared in the transaction record somehow. So it's part of the audit to audit the thinking, which I thought when you said that earlier, it reminded me of dispute resolution. So to actually have that context is great. So I think one thing that stands out to me, though, as a challenge, like we could all agree that this is the process, we need this to be shared, but whenever it terms of data sharing, we see this, course, with transaction data at its core, is standardization and ensuring that this can be done efficiently and at scale. We have wire transaction standards now. We have card network standards. We attempt to standardize this at immense scale across so many different systems and interactions. And I think we're seeing that developing now, and Thailand would love your perspective, Card networks for example are certainly coming out with protocols to try to implement on both the know purchaser the issuing side and the acquiring merchant side and I think merchants certainly have a very interesting role to play here when it comes to kind of mitigating the risk of a dispute and managing risks to consumers when enabling you know agentic payments at the merchant level and I think Visa and MasterCard have both put out their strong opinions there's a lot of standardized protocols being developed to try to add this element of standardization to the messages, to the data sharing, to try to make this process efficient. But I think it seems like the industry broadly agrees, which makes sense that we can't really manage the risks here and we can't effectively resolve any dispute if both parties don't have sufficient context to understand what the intention was and what actually happened.
Tyllen Bicakcic
58:38
Yeah, there were the intention that like, is the, it makes me think about what's the, sometimes you can just think of AI agents as like children or employees, right? And you think, how do you, I'm going to give my card to my kid and they're going go and spend what they want. I actually don't even know how fraud handles this, but it would, I would assume the onus is on me at the end, but. Can I go and dispute it? How would they know? could say, let this go. But then it comes back to, who is Thailand? Who is Thailand giving his card to? Has he ever filed fraud like this before? And then the same thing with an AI. It's safe to say that the top models are getting to the point where they can do what you want. But what model are you using? Maybe you're not using a top model. Maybe who's the vendor that's asking you for this AI agent or that's giving you this AI agent. if it's my own AI agent that I've deployed using open-claw, let's say, you can't blame an open source project, right? So at the end, it's your card. Who's the dispute going to go to? I think if I think like, can you pass prompts in, can you, now it makes sense to me why MasterCard or whoever called it verifiable intense. but I actually still don't know what that means. Partially I need to dive into it, but also like, can the fraud.
lisa
Lisa Durnford
59:57
Mm-hmm.
Tyllen Bicakcic
60:06
You know, like a sardine type C, what is the conversation that led to this AI taking this action? I'm sure in that case, it would make seeing if this is fraud a lot better or not. But it's like, can we standardize getting that information over? Can we standardize which agent acted on whose behalf? Can we standardize authorization for those things? If you can get all of that, it's way easier, right? But as you said, the bottleneck is who's going to sit, you know, how do you enforce that?
lisa
Lisa Durnford
60:35
Yeah, yeah, how do you build that oversight from a centralized perspective? It's always the nacha of payment instructions. Which might be real, yeah, exactly.
Tyllen Bicakcic
60:43
Yeah. Which might have to be the case. Hailey Windham (1:00:49) I think what I was thinking about, especially in regards to disputes with this entire conversation was when I worked disputes, I loved whenever we would get an honest ATM dispute. Cause I was like, I don't have to wait on anybody to provide the documentation. I don't need anybody to provide the video evidence. I controlled it all and it was a click, bing, bang, boom, whatever, pull over all the data I needed. And then I had my decision and the decision was usually.
lisa
Lisa Durnford
60:52
See you.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
61:19
We could find the car that drove up to the ATM, look up the license plate. Does that match who is registered to the account? If not, does that person live in the household of the other? And so then being able to bring back that information to our customer, our member and say, hey, do you know this person? This person is who authorized it. Obviously not sending over the picture in an email because that ends up on Facebook and then you have a whole nother can of worms issue. But I was thinking, you know, as you know, we talked in the beginning about the governance of it, truly that you own it, you bring it on in whatever level your financial institution is ready for. When you have those transactions where I didn't do this or I didn't authorize it, but then you have all of the rationale for why it was accepted, why it was processed. Here's that thinking, here's why we did it or why the AI did it. essentially got your honest ATM dispute where you've got all the proof there that you needed in order to know how to move forward. And so I looked at it and I was more excited because if you dispute this, I already know I have all the data I need to prove that either it was authorized or that someone, you know, exploited a gap or a vulnerability in the system that then we can also pinpoint and fix moving forward.
Tyllen Bicakcic
62:44
Yeah, it's that's the whole hope is it can be a lot easier once we standardize all of this or once, you know, developers or whoever's building the systems are even told how this thing needs to happen in order to get mass market adoption because no one's going to then feed their credit cards or whoever. I think it's also on us to protect the consumers. know a lot of people don't feel safe giving their credit cards to an AI agent. But I know people are feeling safer and safer because as mentioned earlier, you know, our VP of innovation keeps asking us at a bank, why can't this do more? Why can't this do more? The trust is coming from the intelligence that this thing has. So then the why it can't do more is how the governance of these things. But I will say people are going to start doing it and they're already doing it because it's just too valuable at this point. As with everything with AI, think it accelerates everything. And in a way, we're already seeing that with governance.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
63:51
Completely agree. Do you guys have five minutes to finish it up? I know we're over time. Okay, sorry about that.
Tyllen Bicakcic
63:55
I won't. Yeah. No, this is fun. Very, very fun. Feels like we could have that show. This was cool. Yeah, I don't know, a conference or something. Yeah.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
64:00
Ahem.
lisa
Lisa Durnford
64:06
A show.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
64:08
I agree. Yeah, I totally agree. Maybe we'll put this up as a conversation. I'm actually thinking like, I'm going to bring you in on a webinar, Tyllen, if you don't mind. Okay, great. Great. Okay, so moving on, let me write this timestamp down.
Tyllen Bicakcic
64:20
No, that would be awesome. Would love it. Yeah.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
64:29
Right, so think what we are understanding as a part of this conversation is that technology is moving quickly. The governance frameworks often we know take longer, but it doesn't mean that we have to stop innovating and it doesn't mean that we don't need to prepare for this next phase of banking. I'll just call it that. It's the next phase. It's the phase that we technically should already be living in, but unfortunately we have lagged behind, especially as it comes to governance and regulatory frameworks and what are we and how should we and blah blah blah. So, Tyllen, for you, I want to ask, you know, what excites you most about the future of AI in banking?
Tyllen Bicakcic
65:10
One of the things, honestly, it's going to be a sound very cheesy, but we do this every Friday where it's like the people smiles when they use this product or when they use AI in banking. It's just like, I don't think I've smiled with my banker in so long because every time I go there, it's usually like, okay, what mortgage rate are you going to give me? Like, what is, what are you going to tell me that I should be worrying about? What are you going to try to upsell me? Like. This is a way different experience that people have now because it's just more natural. It's, I think that my daughter is gonna grow up in a world where she's definitely gonna talk to her bank and it'll definitely be AI driven and glad her dad had something to do with that. So I think seeing all of these, it's not... I think one of the common misconceptions is that this is for like the next generation, the next generation. It's like, no, this technology does not care on what generation you are. And I have way more smiles from people above 70 years old using this that makes me so happy. That is like, I know we're on the right path and I know this is how people will bank. So our goal every Friday is we show a different smile of a different customer using the product. And we say we're aiming for a billion of these. So that's what we're working. And that's what makes me happy.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
66:36
I love that. And you answered my next question about the misconceptions. I think that's great. It's this is what we can do today. This is possible. And it's not just for the younger bankers that are coming up. It's for you today. And I think that's great. And again, I mean, I'm not a young banker and I found you a use case that I would love to have this at my FI. So yeah, love that and appreciate it. And for Lisa, what advice would you give compliance leaders who are beginning to evaluate AI systems?
lisa
Lisa Durnford
67:08
Yeah, I think if I could zero in on one thing is talk to teams outside of compliance about this. When you're thinking about doing a risk assessment of this new product or whatever it might be, when you're thinking about your enterprise risk program, talk to your product teams, talk to the engineering teams that are building your interfaces, talk to your security teams about authentication and how you're thinking about this. Talk to other folks in the industry. I think we always talk about the risks of managing compliance. in silos and having disparate teams that are kind of all looking at different signals and trying to manage these separately and how that's not the way to go. We have to bring these conversations together. I think that applies so much to this conversation. So risk teams and compliance teams, go meet other teams at the bank and talk to them about how they're thinking about this and what they're seeing and how to build risk systems together to best manage this and that will certainly help build kind of that strongest program to make sure you're actually enabling this and creating a strong business around this functionality and not just thinking about it from a risk perspective.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
68:32
Love that. I love that perspective so much. And then last question, and I'll let you both go. If a bank leader is listening to this episode right now, what is the first conversation they should be having internally, do you think, Tyllen?
Tyllen Bicakcic
68:49
What bank is going to do this that's going to start taking my deposits? Because this is like, quite honestly, like this is kind of, we view this as a very much a go out, acquire customers type thing. And there has never been a stickier technology product than this ever. So who's going to do it in the banking area first? And I will also say once people start having conversations with an AI at your bank, it's very hard for them to leave because all those conversations, all that, it's just like when you talk to Claude versus chat GPT, you probably go to it for different questions based on all the information that has on you beforehand. So it's a real. You know, if I'm the Banksy, I'm talking to our guys and I'm saying, okay, how long do we have until this actually starts becoming a thing? Because we are launching with banks now, we are launching partnerships with LLBs. And I think that it's like, there will be the difference between the types of experiences we see with people who go and handle agentic banking versus not. But yeah, that's my two cents of it.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
69:58
Lisa, any last parting thoughts on this conversation?
lisa
Lisa Durnford
70:03
I think I would just encourage as I always do with compliance teams, encourage them to think about it as how are we going to help our customers use these tools in the best way possible, in the safest way possible. You know, that's the question. And then what does that mean for you and your teams, depending on maybe where you sit in this process, but as a bank, who are the stakeholders that I need around the table to ensure that we're building the best strategy for enabling this and just, you know, encourage everyone to think about this as an opportunity and not necessarily a risk that we just need to mitigate. So really diving in and ensuring that you're asking the right questions and always prioritizing you the customer in how you set this up.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
70:56
agree more. And, Tyllen, any last thoughts?
Tyllen Bicakcic
71:01
No, I love it. I love this whole space. love, mean, there's a reason why I also love working with community and regional banks. Like there's a reason why I live in a small town. understand how important like the personal connection is. I think don't take away that AI takes that away. Like it's not, it's a very personal technology and it is something that improves people's lives dramatically every day. Don't be scared of it. It'll help. People will help people get you comfortable with AI. AI won't help people. That's why we're here. At the end, it's all still a people business.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
71:41
For sure, for sure. So if there's one thing I hope you listeners take away from this conversation, it's this, know, AI is coming into banking whether we're ready or not. The question isn't whether AI will participate in financial services, but the question is, will banks design it responsibly? Because fraud is still human driven. Even when AI is used to commit fraud, it's directed by people. And that means AI designed well, can actually reduce it. So it can enforce the policy consistently, it can monitor behavior in real time, it can operate within guardrails that humans sometimes miss under pressure. We've all seen that. But only if we build it intentionally. Agentic AI isn't about replacing fraud teams at all. It's not about cutting compliance out of the process. It's about embedding trust directly into the system. And so when money moves, especially in real times, governance moves with it. For community banks, isn't something to fear. Like Tyllen said, it's an opportunity to leapfrog legacy thinking and design infrastructure that's safer, smarter, and more defensible. Because if we don't design AI agents inside the banking system with the right controls, someone else will design them outside of it. And that's not the future that any of us want. So we'll part with, stay vigilant, stay informed, and keep moving fraud forward. Thank you guys so much for joining.
lisa
Lisa Durnford
73:04
Thank you.
Tyllen Bicakcic
73:04
Thank you.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
73:06
And before we hang up, I'm just gonna do a quick little welcome back to part two, just in case we split this up into two because it's so long. And I'll just say, guys, welcome back, and then you can just say whatever real quick, if you don't mind. All right, so welcome back to part two. We are very excited. Obviously, we got really deep into the conversation, but we weren't quite finished. So now we're gonna move on into disputes. And then where we are really excited, for where agentic banking is today, where it's going, and where you are ready for it, whether you believe it or not. So, Tyllen, Lisa, welcome back to the show.
Tyllen Bicakcic
73:44
Thank you. Nice to be here.
A blonde woman in a black blazer smiles slightly against a purple background.
Hailey Windham
73:48
Perfect, okay great. Thank you so much guys, I really appreciate it. I'm so sorry for the technical difficulties earlier. I am so embarrassed by that.
Tyllen Bicakcic
73:57
No, no, literally