BioCatch survey reports 80% of bank respondents saw agentic AI attacks

A person using a smartphone banking app, illustrating account access monitored for AI-agent fraud.
Photo by RDNE Stock project on Pexels

Survey points to an emerging control gap

BioCatch's 2026 survey found that 80% of respondents had seen agentic AI used in attacks against their institutions. It gathered responses from 1,440 bank professionals across 25 countries, working in risk-and-compliance, fraud-management and AML. A further 84% ranked AI agents as the sector's most exploitable weakness over the coming year. Some 72% anticipated serious difficulty telling legitimate AI-supported activity from actions that are malicious or manipulated.

BioCatch says its AiRE (Agentic AI Risk Essentials) capability identifies agents operating with an account holder's authority or under a fraudster's direction before funds are transferred. The company's behavioral-biometrics platform analyses signals such as keystrokes, mouse movements, touchscreen behaviour and AI-agent use. BioCatch says AiRE surfaces indicators including:

  • Synthetic inputs and browser permission settings
  • Declarations made by agents and the infrastructure behind them
  • Cloud-hosted agents, including Muse

Broker integrations add another route

In 2026, MetaQuotes and Spotware enabled traders to connect accounts to AI agents through MCP integrations. An FM Intelligence review found Anthropic's Claude integrations at 9 out of 10 brokers. The expansion makes the agent's access path and authority relevant to firms managing account security, not just the agent's trading decisions.

Major UK banks including HSBC, Nationwide, NatWest, Barclays and Lloyds have reportedly blocked AI agents from customer accounts over fraud and liability concerns, Finance Magnates reported. Robinhood takes a different approach: it requires a dedicated Agentic Account for agent use. The company told Finance Magnates that "customers are solely responsible for how the agents are built and any resulting actions the agents take." BioCatch Chief Product Officer Ayelet Eliezer said institutions need to establish whether a session is controlled by a person or an agent, and whether its intent is legitimate or harmful.

What this changes for operators

For broker and bank teams, the operational question is where agent-detection indicators enter the existing decision path. Can the signals be consumed by fraud and risk workflows before a transaction proceeds?

What evidence will the system retain to show whether access was authorised, reviewed, permitted or blocked? BioCatch says its telemetry can inform those choices, but its effectiveness depends on implementation and the platform involved.

Teams assessing agent access should clarify which agents are permitted, how customer authority is established, and how suspicious sessions are escalated. They should also ask how agent-related signals are recorded alongside the resulting account activity.

As agents become better at imitating human interaction, device- and IP-based checks alone may not settle whether a session is genuine. Operators should check how those controls interact with onboarding and existing KYC and fraud processes.

The survey gives no implementation, latency or operating-cost details for AiRE. Those remain questions for institutions evaluating the capability, including how its indicators are delivered to existing systems and what review process follows a risk flag.