AI agents as the banking industry’s greatest vulnerability in next year

South African banking leaders see AI agents as industry’s greatest vulnerability in next year.

New Survey Highlights Growing Concern Over AI-Driven Fraud

New survey of 1,440 fraud and risk leaders across 25 countries finds 96% of South African banking leaders say AI has already increased the sophistication of fraud, eight points above the global average.

JOHANNESBURG (June 10, 2026) — Artificial intelligence is rapidly reshaping the fraud landscape, and South African banking leaders appear among the most concerned globally.

In a new survey of 1,440 fraud management, anti-money laundering (AML), and risk and compliance leaders across 25 countries, including 80 from South Africa, 96% of South African respondents say AI has already increased the sophistication of fraud and scam schemes, while 86% of those surveyed in the country identify AI agents as the industry's most exploitable vulnerability over the next year.

Globally, 80% of institutions reported having already encountered attacks involving agentic AI.

The findings form part of BioCatch's Future of digital trust survey, commissioned by BioCatch, a global leader in behavioural intelligence for fraud and financial crime prevention.

The research explores how financial institutions are responding to evolving fraud threats, the growing influence of artificial intelligence and the role of collaboration in strengthening financial crime prevention, while highlighting the continued rise of fraud and scam activity around the world.

AI Impact

South African respondents report particularly high levels of concern about the impact AI will have on fraud detection and prevention in the years ahead.

Four in five (80%) believe it will be very challenging to distinguish legitimate AI-assisted actions from malicious or manipulated activity, while 64% expect widespread AI-mediated banking to reduce the effectiveness of traditional fraud signals.

“AI is starting to reshape how customers interact with e-commerce sites and financial institutions and will change how criminals execute fraud and other financial crimes,” BioCatch Global Advisory Director Jonathan Frost said.

“As digital interactions continue to grow faster, more automated, and increasingly driven by agents, we must move beyond static identity checks and toward a deeper and immediate understanding of behavior, intent and trust.”

Fraud management and financial crime prevention professionals have witnessed AI specifically amplifying several types of fraud and scams, including deepfake-enabled social engineering, automated phishing, and automated money laundering and transaction fraud.

Rising Fraud

The survey also highlights the growing financial impact of fraud on South African institutions.

Nearly four in five respondents (78%) say fraud attempts are increasing at their organisation, while 79% report rising fraud losses.

Almost half (44%) estimate their institution loses more than $10 million annually to fraud, with nearly a third (29%) reporting losses exceeding $25 million per year.

Beyond institutional losses, respondents indicate that scams continue to have a significant impact on customers.

Forty-five percent estimate their customers lose more than $5 million annually to authorised fraud and scams, while 20% place customer losses above $25 million each year.

Fraud is also becoming faster.

An overwhelming 98% of South African banking leaders say they are very concerned about the increasing speed of fraudulent activity, significantly higher than the global average of 76%.

Today’s criminal networks are unrestrained by the regulatory requirements imposed on legitimate businesses, meaning they evolve quickly, utilising emerging technologies and methods to enhance their abilities.

More than half of those surveyed globally (59%) say criminal enterprises are evolving faster than financial institutions.

The Challenge Of Agentic Banking

The use of agents in retail banking promises to make it even more difficult for financial institutions to differentiate between genuine and fraudulent sessions.

In an agentic future, banks must be able to distinguish between a legitimate user conducting legitimate activity, a fraudster who’s taken over a legitimate user’s account, and a legitimate user under the direction or manipulation of a scammer.

They will also need to discern between genuine users utilising agents to conduct legitimate banking activity and fraudsters utilising agents to conduct fraudulent banking activity.

Four in five (80%) of those surveyed in South Africa say it will be very challenging to distinguish legitimate AI-assisted actions from malicious or manipulated AI activity in a future where AI agents commonly initiate or assist with online banking transactions (vs. 72% globally).

Inter-Bank Intelligence Sharing

Against this backdrop, South African banking leaders are placing increasing value on collaboration and intelligence-sharing.

More than nine in ten South African respondents (91%) believe interbank intelligence-sharing would have a significant positive impact on their ability to prevent fraud and financial crime.

The same proportion say access to real-time intelligence and information sharing on receiving accounts involved in transactions would significantly improve their ability to identify and stop scam attempts that elude traditional defences.

The findings reflect South Africa's longstanding recognition of the value of collective action in tackling organised financial crime, including through industry-wide collaboration initiatives such as SABRIC.

Additional Key Findings From South Africa

South Africa recorded the highest proportion of respondents globally (49%) who believe identity verification and authentication would benefit most from internal agentic AI tools, compared to a global average of 27%.

South African banks appear to be outperforming many global peers in identifying mule accounts earlier in the laundering process, with 39% saying they typically detect mule accounts before receiving any inbound payments, compared to the global average of 20%.

The findings underscore the extent to which South African banking leaders view AI-driven fraud, intelligence-sharing and collaboration as central to the future of fraud prevention and digital trust.