Risk-based authentication is becoming core to financial fraud defence

At a glance

What this is: This analysis argues that financial institutions need risk-based authentication because point-in-time checks miss multi-step fraud and cannot govern what a session is allowed to do once it is inside.

Why it matters: It matters because IAM and fraud teams must move beyond login-centric controls and align identity decisions to session risk, channel context, and evolving customer behaviour across human and agent-mediated activity.

By the numbers:

• 97% of consumers say fraud prevention is a top factor when choosing a financial institution.

Context

Financial institutions often treat authentication as a single decision at login, but fraud now develops across a sequence of interactions that can look benign in isolation. Once the institution has fragmented signals across onboarding, servicing, call centres, and digital channels, static controls struggle to separate legitimate customers from coordinated fraud.

The article is really about the governance gap between identity verification and ongoing trust decisions. In practice, that gap now spans human users and AI agents acting on behalf of customers, which means the control plane has to evaluate not just who authenticated, but what that session should be permitted to do next.

Key questions

Q: What breaks when authentication is limited to a single login decision?

A: Point-in-time authentication misses multi-step fraud because attackers can stay within acceptable thresholds at each event while the overall sequence is malicious. That is why login-only controls often fail against account takeover. The control breaks when session context, channel history, and behavioural drift are not part of the decision.

Q: Why do AI agents complicate customer authentication for banks?

A: AI agents complicate authentication because a successfully authenticated session can still be unsafe if the agent is allowed to perform the wrong action. Banks may accept an agent for read-only tasks but not for transfers or profile changes. The governance problem is permissioning inside the session, not simply proving identity at entry.

Q: How do banks know if accessible authentication is actually working?

A: Look for completion rates across user groups, support-assisted sign-in volume, exception requests, and the number of flows that fail under assistive technology testing. If one browser, device, or access mode consistently produces failures, the control is not robust enough for production use.

Q: When should organisations step up or terminate a session instead of challenging every login?

A: They should do so when risk accumulates across a sequence of actions, not just when one event looks unusual. A new device, silent notifications, contact-detail change, and an unknown transfer target together justify stronger action far more than any one signal alone. The point is to contain the session before money moves.

Technical breakdown

Why point-in-time authentication misses account takeover chains

Traditional authentication answers a narrow question: did the claimant satisfy the checks at a single moment? That works poorly when fraud is distributed across many low-signal actions, such as password reset, contact detail change, and transfer initiation. Risk-based authentication shifts the unit of analysis from a single login to a sequence of events, which is the right model for account takeover. It combines device, behaviour, channel, and history signals so the system can degrade trust as the pattern becomes more suspicious.

Practical implication: teams should move fraud controls from event-by-event approval to sequence-aware decisioning.

How session-aware decisioning changes entitlement control

Session-aware decisioning treats authentication as a living trust state rather than a one-time pass or fail. Each in-session action can increase or decrease risk, and the policy engine can respond by stepping up, constraining entitlements, or ending the session. That is materially different from static MFA because the control is not limited to proving identity at entry. It governs the scope of activity after entry, which is where many fraud patterns actually succeed.

Practical implication: define policy triggers for in-session entitlement restriction, not only for initial login.

Why AI agents create a new permission problem for banking authentication

The article’s Know Your Agent problem is less about detecting an AI label and more about deciding what an authenticated non-human actor may do within a session. An agent may be acceptable for balance checks but not for transfers or contact changes. That introduces a permission model that is dynamic, task-scoped, and context-dependent. Static infrastructure cannot reliably express that boundary if access decisions are frozen at authentication time.

Practical implication: banks need session policies that distinguish permitted read actions from higher-risk write or transfer actions for agents.

Threat narrative

Attacker objective: The objective is to move from initial authenticated access to authorised financial action before the institution recognises the pattern as fraudulent.

1. Entry occurs through compromised credentials, password resets, or new-device logins that appear legitimate at first pass.
2. Credential use is validated once, but the attacker or fraudulent actor then accumulates trust through ordinary-looking interactions across channels.
3. Escalation happens when contact details are changed, notifications are silenced, or transfer workflows are initiated from the same session.
4. Impact is account takeover, fraud loss, and customer trust erosion before the institution can react.

Breaches seen in the wild

• SonicWall SSL VPN account compromises 2025: Attackers used valid credentials to log in to more than 100 SonicWall SSL VPN accounts across 16 environments in October 2025.
• Dropbox Sign breach 2024: A compromised back-end service account gave attackers Dropbox Sign customer data, including API keys, OAuth tokens and MFA information.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Risk-based authentication is becoming the control plane for customer trust, not just a fraud filter. Static login checks no longer match how fraud actually works because attack paths now span onboarding, servicing, and transaction steps. The governance issue is not whether MFA exists, but whether the institution can continuously re-evaluate trust as the session evolves. Practitioners should treat authentication as an ongoing policy decision, not a gate at the door.

Continuous trust profiles expose the real weakness in siloed identity systems: most institutions already collect the signals needed to distinguish legitimate users from fraud, but they fail to carry those signals forward into later decisions. That is an identity governance problem, not a data scarcity problem. When onboarding evidence, behavioural history, and channel context do not inform runtime decisions, every channel becomes a separate trust silo. Practitioners should unify those signals into a single decision fabric.

Know Your Agent is the next governance boundary for financial authentication: the question is no longer only whether the actor is human, but what actions an authenticated non-human or agent-mediated session may perform. That shifts the control problem from verification to permissioning inside the session. The implication for IAM and fraud teams is that step-up rules alone will not define acceptable delegated action.

Session-aware restriction is the practical expression of blast-radius control in banking identity: if risk can rise mid-session, then entitlement scope must be able to shrink mid-session. That reframes authentication from an admission function into a containment function. The practitioner takeaway is that access models must be designed to degrade safely, not merely admit or deny at the first checkpoint.

What this signals

Continuous trust profiling is now a governance requirement for financial identity programmes. Institutions that keep authentication tied to isolated events will continue to miss the sequence-level behaviour that fraud now exploits. The practical shift is toward runtime trust evaluation that can adapt as the session unfolds, rather than after the damage is done.

Permissioning within the session is the new boundary for AI-mediated banking. Once customers delegate actions to agents, the programme has to decide which actions remain safe under delegation and which require tighter controls. That boundary belongs in policy, not in ad hoc operational judgment.

For practitioners

• Map authentication to session risk, not login state Define policies that can change trust level after entry when device, behaviour, or channel signals shift. Use the same decision logic to step up, constrain, or terminate the session when the accumulated pattern looks fraudulent.
• Carry onboarding signals into runtime decisions Preserve KYC, device, behavioural, and contact history signals so later transactions are judged against a customer baseline rather than a one-time check. This prevents each channel from making isolated decisions with partial context.
• Separate read permissions from action permissions for agents Allow low-risk agent activity such as balance lookup, but require tighter policy for payment initiation, profile changes, and wire transfer actions. Make the permitted action set explicit before the session starts.
• Version every authentication decision for auditability Log why a session was stepped up, restricted, or blocked so fraud teams and regulators can trace the exact trust state at each point. Keep the decision trail tied to the signals used at the time.

Key takeaways

• Authentication failures now matter because fraud increasingly unfolds across multiple steps, channels, and decisions rather than a single login.
• The strongest programmes use persistent trust signals to reduce false positives while still catching suspicious patterns earlier in the session.
• Financial institutions need policies that can narrow or end a session as risk rises, especially when AI agents act on behalf of customers.

Key terms

• Risk-Based Authentication: An access model that changes verification requirements based on the estimated risk of the request. It combines identity assurance, device posture, application sensitivity, and contextual signals to decide whether to allow, block, or step up verification before access is granted.
• Session-Aware Decisioning: A control pattern that evaluates risk continuously during a live session instead of only at the point of login. It matters because many fraud and abuse scenarios emerge only after initial access, so policy must be able to react as the session evolves.
• Persistent Trust Profile: A continuously updated identity record that carries forward high-value signals collected across onboarding and later interactions. For financial institutions, it helps runtime controls compare current behaviour against a known baseline rather than treating every event as isolated.
• Know Your Agent: A governance pattern for verifying that an AI agent is the intended actor, not a hijacked or impersonating one. It extends identity assurance beyond login by focusing on runtime behaviour, delegated authority, and the integrity of the agent's actions across connected systems.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.