Financial institutions in parts of APAC cited rising fraud attempts, AI-boosted scam sophistication and agentic AI complicating detection in Jan 2026.

Based on a January 2026 survey* of 1,440 fraud management, financial crime prevention, and risk and compliance professionals at financial institutions in 25 countries on the future of digital trust and AI-driven fraud, some findings on selected respondents’ views of agentic AI and scam risks were shared with the media: Among 340 surveyed decision-makers at financial institutions in the parts of the Asia Pacific region (APAC), the findings reflected how respondents at the time of the survey had viewed AI’s impact on fraud patterns and controls.

First, 86% of surveyed decision-makers from parts of APAC had indicated that they thought AI tools had increased the sophistication of fraud and scam schemes targeting their institutions. A similar 86% had selected survey options indicating a belief that AI agents could become one of the largest exploitable vulnerabilities for the regional banking sector within the following year.

Second, 81% of the same respondents from parts of APAC had responded that fraud attempts at their own institutions were increasing year over year, while 78% had reported year-over-year growth in fraud losses. Within this cohort, 49% had indicated that their institutions were losing more than US$10m  annually to fraud, including 22% who had selected ranges above US$25m, 8% above US$50m, and 2% above US$100m. According to the same regional slice of the survey, 46% had reported that their customers were losing more than US$10m a year to authorized fraud and scams, based on respondents’ self-reporting of internal estimates.

Other findings

Third, 79% of this group of respondents had indicated that their institutions had already encountered attacks involving agentic AI, rather than treating such activity as a purely hypothetical threat. In parallel, 78% had selected survey responses suggesting that, in a future where AI agents routinely initiate or assist banking transactions, they expected it to be very challenging to distinguish legitimate AI-assisted actions from malicious or manipulated activity. Also:

• 67% of surveyed decision-makers in the parts of APAC polled had indicated that their institutions’ current approaches to fraud and financial crime prevention and reimbursement had resulted in a net loss of customers, either through unreimbursed losses or excessive friction.
• 92% had selected options indicating that gaining real-time intelligence on the receiving account in interbank transactions would improve their institution’s ability to recognize and stop scams.
• 93% had indicated that broader real-time interbank intelligence sharing on suspicious activity would have a positive impact on efforts to stop fraud and financial crime across the system.

According to Gadi Mazor, CEO, BioCatch, the firm that commissioned the entire survey, respondents had reported that AI tools were already being used to increase the sophistication and automation of fraud and scams targeting financial institutions, and had cited behavioral and intent analysis, alongside identity verification and network-level intelligence, as among the most critical capabilities for preventing such attacks in an AI‑mediated environment. On that basis, he argued that fraud controls need to “move beyond static identity checks” towards a “deeper … understanding of behavior and intent” in digital interactions.

*Respondents are stated as “manager-level and above professionals within fraud management, financial crime prevention, and risk and compliance at financial institutions in 25 countries”, allocated in the report as parts of North America (United States and Canada, n=260), parts of Asia-Pacific (Australia, India, Indonesia, Singapore, Thailand, n=340), parts of Europe, the Middle East and Africa (the United Kingdom, Germany, France, Italy, Spain, the Netherlands, Belgium, Sweden, Austria, Switzerland, Denmark, Saudi Arabia, UAE, South Africa, n=620) and parts of Latin America (Argentina, Brazil, Colombia, Mexico, n=220) — with a reported margin of error of plus or minus 2.5 percentage points. The country-to-region grouping is inferred from the country list and regional totals published in the methodology rather than explicitly tabulated in the report. No other basic methodological details were provided in the full survey report.