AI fraud is breaking legacy identity controls and SMS OTP

At a glance

What this is: This is an analysis of how AI-enabled fraud is undermining legacy identity verification and why possession-based authentication is emerging as the more resilient control model.

Why it matters: It matters because IAM, fraud, and identity teams need to move away from human-mediated challenges and toward controls that remain verifiable when attackers can clone voices, generate documents, and automate phishing at scale.

By the numbers:

• Over 80% of phishing attacks now target mobile devices.
• More than 60% of internet traffic already comes from mobile devices.

👉 Read IDlayr's analysis of AI fraud, possession factors, and mobile identity

Context

AI-enabled fraud has exposed a basic weakness in identity programmes: controls that depend on a person recognising, entering, or verbally confirming a secret can now be defeated at scale. Knowledge factors such as passwords, PINs, SMS OTPs, and security questions are no longer behaving like durable identity controls because attackers can automate the social engineering around them.

The identity problem is broader than fraud alone. As mobile devices become the primary endpoint for payment, communications, and soon agent-driven transactions, organisations need trust models that bind the user, the device, and the network together without depending on human judgement at the point of challenge.

Key questions

Q: How should security teams replace SMS OTP in high-risk identity flows?

A: Security teams should phase out SMS OTP first in recovery, account takeover prevention, and payment-adjacent journeys, where social engineering has the highest payoff. Replace human-entered codes with possession-based verification tied to a specific device or SIM, and keep the check server-side so the user never transmits a reusable secret.

Q: Why do AI-enabled attacks weaken knowledge-based authentication so quickly?

A: AI lowers the cost of convincing a person to reveal a secret or complete a fraudulent challenge. That means passwords, PINs, security questions, and OTPs are no longer protected only by obscurity. Once attackers can clone voices, generate documents, and relay sessions in real time, knowledge factors stop being a durable boundary.

Q: What breaks when identity verification depends on the user spotting fraud?

A: The control fails when the person being protected becomes the detection layer. Human review is slow, inconsistent, and easy to bypass when attacks look legitimate and arrive continuously. That creates a gap between the speed of the fraud and the speed of the response, which is why silent verification is gaining value.

Q: Who is accountable when AI agents initiate transactions on behalf of users?

A: Accountability should remain with the organisation that authorises the transaction path, but the trust chain must prove which user, device, and agent were involved. Without that binding, it becomes hard to separate legitimate delegated action from hijacked or synthetic activity. That is why transaction provenance needs to be part of identity governance.

Technical breakdown

Why knowledge-based authentication fails under AI fraud

Knowledge-based authentication relies on information a person is supposed to know, such as a password, PIN, OTP, or security answer. AI changes the economics of stealing that information because phishing, voice cloning, and relay kits can now be personalised, continuous, and low cost. The attacker no longer needs to brute force identity. They only need to persuade the user to reveal or relay a secret once. In practice, that makes shared knowledge a fragile control when the adversary can generate convincing, real-time social engineering at scale.

Practical implication: treat passwords and OTPs as insufficient for high-risk journeys, especially where fraud loss or account takeover would be material.

What possession-based authentication changes for mobile identity

Possession-based authentication moves the trust anchor from remembered or observable information to something physically held, such as a device or SIM card. The key technical difference is that verification happens at the device or network layer, where the attacker cannot simply ask the user to reveal the factor. In the model described here, the control is silent, deterministic, and resistant to phishing because there is no code for the user to type and no voice prompt to fake. That makes it structurally different from SMS OTP and other user-mediated checks.

Practical implication: use possession-based controls for step-up authentication and recovery flows where user-interaction itself creates risk.

How agentic commerce alters identity and fraud assumptions

Agent-driven transactions change the identity question from who entered the code to who authorised the action behind the agent. Once software can initiate purchases or financial decisions on behalf of a person, the trust chain must extend across the user, the mobile device, and the application executing the transaction. Without that chain, attackers can hijack sessions, inject false identities, or abuse delegated access in ways that traditional consumer authentication was never designed to detect. This is less about stronger login and more about binding identity to the transaction path itself.

Practical implication: design agent-facing flows so that each transaction can be cryptographically tied back to a verified user and a trusted device context.

NHI Mgmt Group analysis

AI fraud has broken the assumption that a person can safely mediate authentication in real time. Legacy identity programmes were designed for a world where users could distinguish legitimate requests from malicious ones and where challenge responses were slow enough to evaluate. That assumption fails when phishing, voice cloning, and relay automation can complete a fraud path in seconds. The implication is that human-in-the-loop authentication is no longer a reliable control boundary for high-risk access.

Possession factors now represent the more stable trust primitive for consumer and workforce identity. When the factor is physically held and verified at the device or network layer, the attacker loses the easy path of stealing or synthesising a secret. That matters because AI can replicate knowledge and mimic biometrics, but it cannot plausibly manufacture possession of a specific device or SIM card. Practitioners should treat possession-based verification as the control class that best survives AI-enabled impersonation.

Mobile identity is becoming the centre of gravity for both fraud and access governance. The article is right to shift attention from the desktop login to the mobile device, because the phone is already the primary transaction surface for payments, communication, and recovery. As that surface expands, identity controls that only protect the application layer will leave the device and network relationship exposed. Security teams need to align IAM, fraud, and mobile trust decisions around the same identity binding model.

Agentic commerce creates a new trust problem: proving who stands behind the action, not just who started the session. Once an AI agent can act financially on a user’s behalf, access governance must extend beyond session authentication into transaction provenance. That is where traditional auth patterns become thin, because they do not reliably preserve accountability across delegated execution. The field now needs identity models that preserve provenance from person to device to agent to transaction.

From our research:

• The average estimated time to remediate a leaked secret is 27 days, despite 75% of organisations expressing strong confidence in their secrets management capabilities, according to The State of Secrets in AppSec.
• Only 44% of developers are reported to follow security best practices for secrets management, which helps explain why identity controls that rely on human discipline continue to fail under pressure.
• For a broader NHI governance lens, see Ultimate Guide to NHIs , Key Challenges and Risks for the operational patterns that make secrets hard to contain.

What this signals

As AI-driven fraud shifts more identity risk onto mobile and transaction flows, the programme question is no longer whether OTP can be hardened. It is whether identity assurance can survive when the attacker controls the conversation. That pushes teams toward device-bound trust, recovery hardening, and tighter alignment between IAM and fraud operations.

Credential replay debt: controls that still depend on reusable secrets are accumulating operational debt faster than most teams can remediate it. The deeper issue is not just exposed credentials, but the lag between compromise, detection, and revocation. Teams should expect pressure to shorten that gap as AI makes social engineering cheaper and more scalable.

If your identity stack still treats mobile as a secondary channel, your trust model is already behind the way users, payments, and delegated actions now work. Mature programmes will start measuring how often a user-mediated step is the actual point of failure, then replace those steps with verification that does not rely on human judgement under attack.

For practitioners

• Replace SMS OTP on high-risk journeys Move password resets, account recovery, payout approval, and sensitive profile changes away from SMS OTP and other user-entered secrets. Prioritise flows where phishing or relay attacks would create immediate financial or identity loss.
• Bind authentication to device and network context Use possession-based checks that verify the SIM or device at the network layer so the control does not depend on a user reading or typing a code. This reduces exposure to deepfake voice calls, real-time phishing, and social engineering.
• Redesign recovery and step-up controls Treat recovery, SIM swap handling, and step-up challenges as the highest-risk identity moments. Require controls that remain valid even when an attacker can imitate the user convincingly or operate faster than a human can respond.
• Extend trust into agent-driven transactions If your business is exploring agent-based commerce, make the transaction verifiable back to a known user and trusted device before the agent can execute a financial action. This helps preserve accountability when software acts on behalf of a person.

Key takeaways

• AI has turned human-mediated authentication into a weak point because secrets, voices, documents, and challenge flows can now be copied or simulated at scale.
• Possession-based controls matter because they verify something the attacker cannot easily fabricate, especially when the check happens at the device or network layer.
• The next identity governance problem is not just login security, but preserving trust and accountability as mobile and agent-driven transactions become normal.

Key terms

• Possession Factor: A possession factor is something a user physically holds or controls that can be used to verify identity. In practice, this includes a device or SIM card. The control is stronger than knowledge-based checks because it does not depend on secrets that can be guessed, phished, or copied.
• Knowledge-Based Authentication: Knowledge-based authentication verifies identity using information the user is expected to know, such as a password, PIN, OTP, or security question. It is brittle under AI-enabled fraud because attackers can steal, synthesize, or socially engineer those answers at scale.
• Silent Network Authentication: Silent Network Authentication is a device or network-layer verification method that confirms possession without requiring the user to type a code or complete a challenge. It reduces phishing exposure because the trust decision happens behind the scenes, not through a reusable secret.
• Agentic Commerce: Agentic commerce is the use of AI agents to make purchases or financial decisions on a person’s behalf. It changes identity governance because organisations must prove not only who started the session, but who is accountable for each delegated action and transaction.

What's in the full article

IDlayr's full article covers the operational detail this post intentionally leaves for the source:

• Practical explanation of silent network authentication and how the verification flow works end to end
• Direct comparison of SMS OTP, passkeys, and SIM-based possession factors for fraud resistance
• Specific examples of how mobile trust can be extended into agentic commerce and transaction approval
• FAQ-level detail on when device-bound verification is stronger than knowledge-based or biometric checks

👉 The full IDlayr post covers the mobile trust chain, agentic commerce risk, and silent authentication details.

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 building or maturing an IAM programme, it is worth exploring.