Connect with us

Banking & Payments

HKMA planning sandbox for generative AI in banking

HKMA’s Arthur Yuen says the Silicon Valley Bank failure holds alarming lessons about digitizing finance.



Arthur Yuen, HKMA

Arthur Yuen, deputy CEO of Hong Kong Monetary Authority, says the territory’s central bank is preparing to open a regulatory sandbox focused on how financial institutions may use generative artificial intelligence.

Speaking at a conference organized by The Asian Banker, Yuen expressed alarm at the bank failures of March 2023, which demonstrated new risks to financial stability arising from digital innovation.

HKMA is working with a tech partner, which Yuen didn’t name, to help it understand the programs being trialed within the sandbox.

A sandbox regime allows banks or others to experiment with new business models or capabilities under the promise of supervisory leniency. It’s a way for central banks to keep a close eye on innovation, while giving banks comfort that they won’t be unduly punished when newfangled tools go awry. Once the central bank is satisfied the banks have a strong culture of risk management around what’s being tested, the products are allowed to be fully deployed.

The dramatic arrival of generative AI using large-language models based on transformer mechanics in computing holds great promise for further transforming banking. But from a regulator’s point of view, the risks are substantial.

The run on SVB

Yuen pointed to the March, 2023 collapse of three banks in the US (Silicon Valley Bank, Signature Bank and Silvergate Bank), as well as Switzerland’s shutting down Credit Suisse, as harbingers of new risks to financial stability.

Unlike previous bank failures, these events were defined by technology elements, Yuen said. Firstly, SVB and others were both digitally savvy and designed to cater to tech companies. Secondly, their customers were heavily digitalized. Thirdly, their downfall was accelerated by the rapid dissemination of rumor and fear over social-media networks.

What these factors add up to is speed. Yuen notes that SVB suffered $140 billion of deposit outflows in two days; and it took only two more days for the bank to fail. He compared that to the worst deposit flight from a commercial bank in the 2008 financial crisis: US lender Wachovia lost $10 billion over eight days.

New risks

For HKMA, the lesson is that liquidity management risk is now as important as how banks manage market or counterparty risks.

It means that commercial banks must sharpen their pencils when it comes to liquidity, operational resilience, and understanding how such failures impact their customers – who can now shift their funds with just a few clicks on their mobile phones.

“Is there still a viable business model for banks?” Yuen postulated. “Yes, if we can ensure that we enhance liquidity risk management.”

That means assuming that innovation will continue to impact this risk. He also called on banks to monitor social media. HKMA has found that tracing certain keywords on Twitter and other platforms show how the SVB collapse was narrated in real time. That’s a backtest, but it suggests such monitoring could help bankers and regulators appreciate the scale of an unfolding drama.

He also suggested banks “pre-position” collateral to ensure central bank support should something go wrong.

The onus for grappling with this risk is shared among commercial banks and their regulators. Central banks must be ensure they can make contingent funding available much more quickly.

(Yuen was too polite to mention that SVB’s failure was in part due to just missing a 4pm cutoff to access the Federal Reserve’s discount window, which left the authority unable to serve as lender of last resort due to its own operational rigidities.)

Operational onus

To date, no bank has ever failed due to operational failures. But Yuen wondered if that could change. Already, IT failures are becoming hugely disruptive and expensive to banks. Bank systems are getting more difficult to manage as banks try new technologies.

Yuen notes that banks are now partnering with fintechs more to learn how to deploy new tech, such as crypto, DeFi, and tokenization. Bankers these days talk proudly of being less obsessed with building everything in-house, and partnering with tech companies. But from the regulator’s perspective, this is a risk, as banks are now dependent upon external vendors. Such relationships must be part of the bank’s risk-management process, Yuen said.

Fraud and cyber crime is another increasing risk. This is different to the factors behind a run on a bank. But it is related as digitalization is also becoming a handy tool for criminals. GenAI and deepfakes are already being deployed by villains to fool CFOs into making false transactions: UK engineering firm Arup’s Hong Kong CFO fell victim to a sophisticated, multi-party deepfake psyop that convinced him to transfer HK$200 million in February.

Yuen noted that HKMA received over 1,200 complaints related to cyber fraud in 2023, double the number in 2022. He is worried that losses could become big enough to threaten a bank, and even the banking system. Banks and commercial interests must collaborate more tightly with regulators and law enforcement to counter these threats, he said.

Innovation agenda

Yeun was eager to show HKMA also works to encourage innovation. It is a big user of regtech tools, has backed the establishment of Commercial Data Interchange (a data credit scoring service for smaller companies), and the introduction of licensed digital banks.

He acknowledged that distributed-ledger technology is going to play a significant role in financial services, and urged banks to be ready to interweave DLT into their overall operations. HKMA’s initiatives include Project Ensemble, a grouping of banks and fintechs aimed at providing a layer of interoperability for tokenized deposits or stablecoins.

Yuen said more details about the genAI sandbox will be revealed later this year, noting that it should be ready to admit banks before the end of 2024.

DigFin direct!

Register to receive DigFin's newsletter

  • Hauptseite
  • Grocery Gourmet Food
  • HKMA planning sandbox for generative AI in banking