Banks have a deepfake problem

The financial industry has a Taylor Swift problem – as in the Swift deepfake problem (allegedly endorsing Trump and, more graphically, here). Australia’s CommBank CEO Matt Comyn’s video double ended up in an investment scam. Bank boards are gaming out worst-case AI scenarios involving deepfakes, the Wall Street Journal reported last week.

Banks are acting on these concerns by staffing up. The number of AI risk-related jobs of the 50 Evident AI Index banks is up 60% since September of last year, according to new Evident research. They’re talking more about it as well. Over the past 12 months, 38 of the 50 Evident AI Index banks reported on the threats to their business from the new technology, up from 27, according to our research.  

In their assessments of risk in the annual reports, banks tend to break them out into three categories:

• Operational: The risk of using an AI tool that’s inaccurate or that uses biased data can lead to bad business decisions being made.
• Third-party: The risk of a vendor-supplied AI model that affects every bank that uses it (as happened with the Crowdstrike outage last month).
• Cybersecurity: The risks that AI tools are vulnerable to new methods of hacking, compounded by fears that employees using AI tools without oversight could leak sensitive information.

Why this added focus now? For sure, bank supervisors in the U.S. and their counterparts in the U.K. and EU are flagging potential problems with AI in the financial industry. Compliance generally follows those signals. Deepfakes are an urgent threat to reputations and potentially the bottom line.

But across the board, we’re seeing that banks are focusing more on how to adapt existing governance systems to AI specific risks and correspondingly staffing up on risk talent. The next big move will likely be toward Agentic AI, which will see the technology act more autonomously – which clearly carries great promise… and risk.

For more on RAI and risk, check out the Transparency section of the 2023 Evident AI Index Key Findings Report.  Our Evident AI Index updates for 2024 on October 17 – a date for your diaries.

Ash Fontana is an investor in AI and author of The AI-First Company: How to Compete and Win with Artificial Intelligence. We sat down with him to discuss where banks, and businesses, are on the path to using this technology. The following conversation has been edited and shortened for clarity.

In the three years since you published The AI-First Company, what’s changed?

Generative AI tools have made it easy to experiment, which means a lot more people are coming up with use cases. A huge amount of companies are getting value out of AI – we can’t even pay for something or look at a map without AI being involved. We use so much AI every single day that people don’t know about, but it’s absolutely everywhere.

Has there been a proliferation of AI-first companies?

No. Building an AI-first company is as hard as ever. Just tacking a chatbot onto existing services doesn't make you an AI-first company. You’ve got to put AI at the start of every decision, whether it's about people, policy, process, products or pricing. Lots of companies are experimenting, but there's a long way to go before they are really AI-first.

How could legacy businesses like banks accelerate their AI adoption efforts?

Everyone in the company has to be able to experiment with AI. You have to enable every single person to do something. That can be as simple as giving everyone access to Copilot. A company needs to experiment with AI on a small scale across every single problem to really see the potential of AI and transform itself.

Bank CEOs face a lot of questions about the size of their AI investments. What can they do to bring investors, employees and customers along with them?

If you just do small, ad hoc, piecemeal-type things, it’s going to be hard to get people to believe that the investment will pay off. You’ve got to present a complete strategy, and really hold yourself to return on the investment. It's very difficult, but you can do it. If you don't, expect incredulity from your stakeholders.

Join us for our upcoming Evident AI Leadership Roundtable: How can banks translate AI strategy into an effective communications plan?

Our Leadership Roundtable offers a unique opportunity to learn from industry practitioners and subject matter experts working at some of the world’s leading banks. Attendees will learn how to translate AI strategy into an effective communication plan – and ask critical questions on this topic.

In this week’s Use Case Corner, we look at three of the most interesting tools unveiled in the past month.

And ICYMI: Check out our special edition on use cases in the last Brief.

#1 Generative AI virtual assistant

Use Case: Gen AI Agent Servicing Tool
Vendor: n/a
Bank: Capital One

Why it’s interesting: Capital One is adding a lot of Gen AI talent, and we’re starting to see what they’ve been working on. The Agent Servicing tool uses generative AI to provide Capital One’s customer service agents with accurate information to respond to customer inquiries.

Potential ROI → faster client resolution time, improved customer satisfaction, higher volume of client queries answered
Reported ROI → over 95% of search results found to be highly relevant

#2 Generative AI Employee Chatbot

Use Case: Ask Archie
Vendor: n/a
Bank: NatWest Group

Why it’s interesting: Ask Archie is a Gen AI-enabled chatbot that answers employee HR-related questions. Graham Smith, NatWest Group’s head of data science & innovation, says they’re looking to reuse the technology developed and the design in other use cases across the bank.

Potential ROI → decreased query resolution time, increased employee satisfaction
Reported ROI → n/a

#3 AI-enabled Cashflow Forecasting

Use Case: Cash Forecasting
Vendor: Kyriba
Bank: US Bank

Why it’s interesting: US Bank is partnering with Kyriba to introduce a cashflow forecasting tool “by the end of the year or early 2025.” It uses AI to help businesses predict their future cash flow by analyzing past financial data. Earlier this year, JPMC’s Cash Flow Intelligence tool reportedly saved manual work by “almost 90%.”

Potential ROI → increased efficiency, increased customer satisfaction
Reported ROI → n/a

Have feedback on or ideas about use cases? Let us know at [email protected].

"If there’s any setup where you can combine the unpredictable but really powerful outputs of AI with independent verification to filter out the rubbish and keep the good stuff, then you start really seeing lots and lots of potential applications."

- Mathematician and Fields Medalist Terence Tao, speaking at Oxford University, 7 August

BBVA expanded its “AI Factory” research center to Mexico and Turkey. The Spanish bank also disclosed information about some of the AI algorithms they use in their consumer products.

CIBC is filling more than 200 data & AI roles, driven by an expanded partnership with Canada’s Vector Institute, as well as joining a cohort of Canadian companies working on responsible AI development.

The Hong Kong Monetary Authority announced a Gen AI sandbox, a controlled environment that lets banks experiment with use cases without fear of regulatory backlash.

ANZ teamed up with Microsoft to launch the AI Immersion Centre, a dedicated facility that will train 3,000 senior employees in generative AI.

BNY’s AI assistant Eliza, built on models from Google, OpenAI and Meta, is used by one in four employees across the bank.

Truist brought on Steve Hagerman, previously CIO of consumer technology at Wells Fargo, as group CIO. The bank also hired Chandra Kapireddy as their new head of Gen AI. He was head of AI products, CCB Risk for JPMC.

Bridget Engle, ex-CTO at BNY, is moving to Wells Fargo as Head of Technology – with current CIO Tracey Kerrins shifting to head up a new generative AI team. Sidenote: Wells Fargo is a top poacher of AI talent.

ING appointed former UniCredit Daniele Tonella as their new CTO.

Lloyds appointed Rohit Dhawan, ex-AWS AI exec, as group director of AI and analytics.

Do you know or run an event that you think should be featured? Let us know at [email protected].