The new offering, Agentic AI Risk Essentials (AiRE), allows financial institutions to identify AI agents in a banking session before any funds are moved. This applies whether the agents are acting for accountholders or being directed by bad actors. Institutions using the capability also receive indicators that flag agentic activity that may carry risk. The launch extends BioCatch’s behavioural approach to fraud prevention to sessions where the entity on the other side of the screen may not be human.
Ayelet Eliezer, Chief Product Officer at BioCatch, said banks may take different stances on agent access. Some may block agents entirely, some may permit only pre-approved agents, and others may open their platforms to any agent. According to the official, in each case institutions must first establish whether a session is controlled by a human or an agent, and whether that entity’s intent is genuine or malicious. Eliezer also noted that AI capabilities are developing quickly for consumers, institutions, and criminals alike.
How the detection works
BioCatch is offering AiRE through existing customer integrations, so current clients can use it within their present set-up. The capability relies on continuous session monitoring. This lets the company both identify agents and assess how they behave, and banks can use that information to build policies on whether a given agent can be trusted.
AiRE signals cover behavioural anomalies, unusual interaction patterns such as synthetic inputs, the browser permission profile, agent infrastructure, and agent declaration. Detection also extends to cloud-based agents, including Muse.
The capability builds on BioCatch’s existing model. That model continuously collects more than 3,000 anonymised data points as people use their digital banking platforms, including keystroke and mouse activity, touchscreen behaviour, AI agent usage, and jailbroken devices. Its AI and machine learning models use these inputs to assess user intent, as well as signs of coercion or manipulation, throughout each session. Insights from across the company’s network of institutions feed into its real-time risk scoring.
Data from BioCatch customer sessions shows that close to half of all detected agentic interactions took place at account opening, while the rest were spread across logins, account management, and transactions. The distribution indicates that agent activity is not confined to payments and reaches into onboarding and account servicing.
Industry context
The launch follows survey data pointing to concerns on both sides of the banking relationship. A recent survey of US consumers found that 80% believe their bank should be able to tell whether it is dealing with a customer or that customer’s agent. BioCatch’s own 2026 report, The future of digital trust, surveyed 1,440 banking leaders worldwide. Of those, 80% said their institution had already encountered attacks using agentic AI, and 84% expect agents to rank first among the industry’s exploitable vulnerabilities over the next year.
