Moving Beyond Knowledge-Based Authentication: How to Protect Against Account Takeover in the Age of Generative AI

Knowledge‑based authentication (KBA) verifies a person by asking for information only the account holder should know — a mother’s maiden name, a first car, the street they grew up on. Static KBA uses answers the user chose in advance. Dynamic KBA generates questions in real time from credit bureau or transaction records. Both are now weak, because the answers are no longer secret and generative AI supplies them instantly.

Knowledge‑based authentication (KBA) has been the backbone of identity verification in call centers and customer service for decades. When customers forget passwords or request account changes, agents ask questions such as “What street did you grow up on?” or “What was your first car?” The assumption is that only the legitimate account holder knows the answers. Over time this method became attractive because it requires no specialized hardware and seems easy to implement and familiar to customers. Unfortunately, KBA depends on two fragile assumptions: (1) the personal data used in questions is secret and (2) human memory limits prevent instant recall. Both assumptions are now false.

Why KBA Is No Longer Adequate

Personal data is no longer private

Static KBA relies on pre‑set security questions such as mother’s maiden name, favorite food or first pet. Once this information is compromised through a data breach or social engineering, it remains compromised indefinitely because the answers do not change. Attackers can mine social media posts, open‑source intelligence and commercial databases to compile a dossier of personal information. A 2021 HelpNet Security article noted that even casual quizzes on social networks— “What was your first car?”—collect answers that can later be abused. Worse, many users reuse the same security answers across multiple accounts, enabling attackers to breach several accounts once they learn a single answer.

Dynamic KBA generates questions on the fly from credit‑bureau or transaction data (e.g., “Which bank issued your mortgage?”). While this improves unpredictability, it still relies on data stored in databases that determined attackers or AI systems can query instantly. The vulnerability is not just that the data is available, but that AI agents can access it faster than humans can think. The article “What Is Knowledge‑Based Authentication? And Why AI Just Broke It” explains how generative AI customers answer questions without hesitation, eliminating the behavioral cues (pauses, self‑corrections, forgetting) that human agents use to detect fraud.

Human memory is unreliable

KBA’s reliance on human memory has become a liability. Users frequently forget what answers they provided years ago. A Google study of its own account-recovery data found that only 47% of users could recall their answer to “Favorite food?” a year later, and that a single guess would match 19.7% of English-speaking users’ answers (Google, 2015). This poor recall frustrates users and leads them to choose easily remembered (and easily guessed) answers such as names or dates. The result is a negative user experience and an increase in costly help‑desk calls for password resets.

Attackers have automated KBA exploitation

Advances in AI have destroyed the friction that made KBA work. AI agents can now answer KBA questions with machine precision in milliseconds (CX Today, 2026). Voice deepfakes can mimic executives or customers, bypassing manual voice‑recognition heuristics. Pindrop measured a rise of more than 1,300% in deepfake fraud attempts during 2024, from about one a month to seven a day, across the 1.2 billion calls it analyzed (Pindrop, Voice Intelligence & Security Report, 2025). As Trusona’s founder Ori Eisen notes, generative AI has evolved from simple phishing into sophisticated attacks on the account‑recovery process, using ID spoofing and deepfakes that simulate voices and documents.

Standards bodies and regulators advise against KBA

Government and industry guidance has moved away from KBA. The U.S. National Institute of Standards and Technology (NIST) removed pre‑registered knowledge tokens from its Digital Identity Guidelines, recognizing that they are “special cases of (often very weak) passwords”. The updated NIST SP 800‑63‑4 explicitly states that knowledge‑based authentication does not constitute an acceptable secret for digital authentication. Even earlier versions of the guidelines emphasized this removal. The U.S. Federal Financial Institutions Examination Council (FFIEC) notes that reliable verification methods “generally do not depend solely on knowledge-based questions to verify identity” (FFIEC, 2021). Industry articles from OneLogin and CX Today highlight that KBA has become a weak factor because the personal information it uses is widely available.

The Rising Cost of Account Takeover Fraud

Account takeover (ATO) occurs when an attacker gains control of a legitimate account, often by hijacking the account‑recovery process. In an Abnormal Security survey of about 300 security professionals, 83% of organizations had experienced at least one account takeover in the past year (Abnormal Security, 2024). The FBI’s Internet Crime Complaint Center logged about 4,700 complaints and $359.7 million in reported losses in 2025 for account takeover fraud through impersonation of financial institution support (FBI IC3, 2025 Annual Report). Across Sift’s network, the ATO attack rate rose to 2.5% in Q2 2025, and attacks in fintech and finance surged 122% year over year (Sift, 2025). Generative‑AI deepfakes exacerbate this trend. Sift measured a 427% increase in account takeover attacks across its network in Q1 2023 compared with all of 2022 (Sift, 2023). AI‑driven deepfakes of voices, documents and selfies have expanded the attack surface for social engineering.

In call centers, the situation is urgent. A poll Neustar ran during a 2021 American Banker webinar found that 63% of banks still relied wholly on KBA questions to authenticate customers (Help Net Security, 2021). Attackers go after these call centers because most large enterprises outsource customer service to third-party contact centers, and hacker rings increasingly target the people who staff them (Help Net Security, 2021). This vulnerability is a key reason that organizations need to modernize their identity verification processes.

The Shift to Risk‑Based Authentication

Security professionals recognize that no single factor can secure accounts. Risk‑based authentication adapts the rigor of verification based on context, combining multiple factors to match the risk of the transaction. Under the FFIEC guidance, authentication controls are applied “commensurate with the increasing risk level associated with a transaction,” with stronger controls for high-risk activity such as a customer initiating a payment (FFIEC, 2021). NIST’s digital identity framework uses Authentication Assurance Levels (AAL) to indicate the strength of the authentication process, with AAL2 and AAL3 requiring multiple factors. However, choosing an approach and building the infrastructure is challenging for many organizations.

Modern identity verification solutions replace or supplement KBA with document verification, biometric authentication (fingerprint, facial recognition, voice recognition), device intelligence, SIM‑swap detection, and behavioral analytics. OneLogin’s analysis of KBA pitfalls recommends moving to multi‑factor authentication (MFA), biometric authentication and behavioral biometrics. These methods are harder to spoof and do not rely on secrets that remain static. Yet even MFA is insufficient when it still includes a password or KBA as the first factor—phishing and SIM‑swap attacks can bypass SMS codes.

Enter ATO Protect: A Modern Identity‑Proofing Solution

Trusona’s ATO Protect is one example of that shift, from static KBA to risk‑based identity verification. It runs no selfie or liveness check, because generative AI can now defeat those methods. Launched in early 2024, ATO Protect was created to secure the account‑recovery process, which is often disconnected from security measures used at account creation. The platform aims to “know who is really on the other end of a digital transaction,” addressing both fraud risk and customer experience.

How ATO Protect Works

ATO Protect uses multiple layers of verification to reduce fraud without requiring complex integration. Key features include:

Deepfake‑resistant identity verification. ATO Protect verifies the caller by reading a government-issued ID and checking it against the source that issued it — in the US, the state DMV through the AAMVA network — alongside mobile carrier and LexisNexis records. It catches forged IDs, synthetic identities and GenAI deepfakes in the same step, and because the caller’s voice is never the evidence, a perfect clone gets a fraudster nothing. According to Trusona, the solution detects SIM swaps and port-outs and blocks man-in-the-middle attacks.

Real‑time risk assessment. The platform requests customers to corroborate their identity and then validates the submitted data with authoritative sources and device behavior. Device intelligence adds signals from the person’s own mobile device, where the check runs. SIM swap and port-out detection flags a number that recently moved to a new SIM or carrier. If risk is high, ATO Protect provides tools for step-up verification before the account is recovered (Trusona, 2024).

No‑integration deployment. Trusona emphasizes that ATO Protect can be deployed as a zero‑integration solution or via an IT service management (ITSM) API. This design allows organizations to trial the service quickly without changing existing infrastructure. For developers, the ATO Protect API works with systems like ServiceNow, Zendesk and Jira.

Passkey support after verification. Once a caller has been fully verified at the help desk, they can set a passkey in their mobile browser, pairing strong identity proofing with phishing‑resistant authentication. Passkeys are a secondary feature of the platform rather than the core product. The verified recovery call is simply the natural moment to retire a password.

Scalability and geographic coverage. ATO Protect supports worldwide scanning of government documents, multiple phone networks and thousands of document types. This breadth makes the solution suitable for multinational corporations.

Why ATO Protect Is Different from Traditional KBA

The core difference is that ATO Protect verifies a real‑world identity rather than knowledge of past events. Checking a government-issued ID against the authority that issued it gives high assurance that the caller is the account owner. It also uses dynamic risk assessment; low‑risk events receive minimal friction, while high‑risk requests require stronger verification. For example, a bank asked to move a significant sum after a change of address would want a higher level of assurance before approving (Trusona, 2024). This aligns with the contextual authentication recommended by risk‑based frameworks.

ATO Protect addresses the three major problems with AI‑driven KBA failure: (1) it recognizes the identity rather than the memory; (2) it moves the check off the call and onto the caller’s own mobile device, so a convincing voice changes nothing; and (3) it lets a team step up to stronger identity checks before authorizing a high-risk request. As Trusona put it at launch, “the ability to deploy increasingly complex identity verification depending on the level of risk the transaction or event presents sets ATO Protect apart from current solutions” (Trusona, 2024).

Case Studies and Adoption

Organizations across industries have adopted ATO Protect. Testimonials on Trusona’s website include financial firms, law practices and venture capital partners who use the solution to verify clients and secure sensitive transactions. For instance, Exeter Financial uses ATO Protect to verify clients and ensure asset security. Kleiner Perkins uses ATO Protect across the organization to confirm the identities of callers. Ballistic Ventures pairs ATO Protect identity verification with verbal protocols to secure wire transfers. In each case the check establishes who the person is, without asking them to recall anything.

Migrating From KBA to ATO Protect

Switching from KBA to a modern solution like ATO Protect involves more than purchasing software; it requires operational and cultural changes. Below are recommendations for organizations planning a migration.

• Assess risk and identify vulnerable workflows

Begin by mapping the call‑center and digital workflows where KBA is currently used: password resets, account recoveries, high‑value transactions, new account creation. Determine which transactions pose the highest risk of account takeover. Sift’s data shows that fintech and finance experience the highest surge in ATO attacks, so these sectors require strong identity proofing.

• Secure executive and legal buy‑in

Investing in identity verification may raise privacy and regulatory questions. Involve compliance teams early to ensure the solution meets privacy laws. Because this kind of verification handles government ID data, organizations should evaluate data handling policies and ensure suppliers hold current attestations (e.g., SOC 2). Trusona stores no PII and is SOC 2 audited. The cost of fraud is what justifies the investment, so put a number on it.

• Pilot with a no‑integration deployment

ATO Protect offers a no‑integration option that allows organizations to test the service quickly. Start with a specific use case, such as password resets or wire‑transfer confirmations, and measure improvements in fraud reduction and user experience. Evaluate metrics such as average call handling time, successful verifications and customer satisfaction.

• Train agents and update policies

The success of a risk‑based system depends on well‑trained agents. Staff should understand how to initiate the ATO Protect workflow, interpret risk scores and proceed with step‑up verification when necessary. Policies must also cover how to handle AI callers who present valid authorization tokens, aligning with the permission‑based authentication model advocated by CX Today.

• Layer with passkeys and device‑based MFA

Document verification is only one piece of the puzzle. For ongoing interactions, encourage customers and employees to register passkeys or device‑based MFA after they have been fully verified. Passkeys eliminate password reuse and are resilient to phishing. A recovery call verified with ATO Protect is a good time to set one up.

• Monitor and adapt

Threats evolve. ATO Protect’s features layer into an existing risk decision matrix (Trusona, 2024). Regularly review fraud trends, update document verification capabilities and integrate with other security controls (e.g., behavioral analytics, transaction monitoring) to stay ahead of attackers.

Conclusion

KBA served a purpose in an era when personal information was hard to obtain and generative AI did not exist. Today, it is an inadequate and risky authentication method. Attackers and legitimate AI agents alike can answer KBA questions flawlessly, removing the friction that once helped distinguish genuine users from fraudsters. NIST’s guidelines state plainly that KBA “does not constitute an acceptable secret for digital authentication” (NIST SP 800-63-4, 2025), and industry data shows that ATO fraud is increasing dramatically. Organizations still relying on KBA risk financial losses, reputational damage and regulatory scrutiny.

Modern identity verification must confirm who is calling, not what they know. Solutions like Trusona’s ATO Protect use document scanning, device intelligence and real‑time risk analysis to verify the real‑world identity behind a high-risk request. By layering these capabilities with passkeys and risk‑based policies, organizations can mitigate account takeover fraud while delivering a smoother user experience. Knowledge‑based authentication depended on two things: answers that stayed secret, and attackers who could not recall them instantly. Breaches removed the first. Generative AI removed the second.

Frequently asked questions

What is knowledge-based authentication?

Knowledge-based authentication verifies a person by asking for information only the account holder should know, such as a mother’s maiden name or the street they grew up on. Static KBA uses answers the user registered in advance. Dynamic KBA generates questions in real time from credit bureau or transaction records.

What is the difference between static and dynamic KBA?

Static KBA draws on answers the user chose earlier, so once an answer leaks it stays compromised — the answers never change. Dynamic KBA builds questions on the fly from records the user never selected, which makes them harder to anticipate, but they remain answerable by anyone able to query the same data.

Why is knowledge-based authentication no longer secure?

It depends on two assumptions that no longer hold: that the answers are secret, and that only a genuine person can recall them quickly. Breach data and public profiles made the answers findable, and generative AI supplies them without hesitation, removing the pauses and self-corrections an agent once read as a warning sign.

Does NIST allow knowledge-based authentication?

No. NIST withdrew pre-registered knowledge tokens, sometimes called security questions or KBA, from its Digital Identity Guidelines on the grounds that they “often rely on information that is private but not secret” (NIST, SP 800-63-3 Implementation Resources). NIST also prohibits hints in prompts for memorized secrets, citing similar security concerns and the possible use of hints as a work-around to support security questions.

What can replace knowledge-based authentication?

Verification against sources the caller does not control. Checking a government-issued ID against the authority that issued it — in the US, state DMV records over the AAMVA network — establishes who someone is without asking them to recall anything. SIM swap and port-out detection, and man-in-the-middle detection, add context from the carrier and the live session rather than from the caller.

Is KBA still used in call centers?

Yes, because it is inexpensive and needs no hardware on the caller’s side. That is also why attackers prepare for it. Any team still using knowledge questions to approve password resets, account recovery, or changes that move money is relying on information that has already been published somewhere.