Reflectiz AI Agent Identifies OTP Bypass on Insurer Site
Reflectiz said its AI pentesting agent identified and validated a critical authentication bypass in the production login flow of a major insurer with millions of users. During an authorised assessment, the agent requested a one-time passcode, redirected it to an email inbox it controlled, completed the login and then stopped under its guardrails.
The company reported the finding to the insurer. Reflectiz did not name the organisation, and the account state shown in its account of the test was recreated and anonymised.
An OTP destination controlled by the request
The issue was in how the login request selected delivery of the one-time passcode. The request included fields for both the delivery channel and the destination address. Ordinarily, a code was sent by text message to the phone number associated with the account.
Reflectiz said that changing the channel to email caused the server to send the code to the address supplied in the request, rather than verifying that the address was registered to the account. In effect, the server did not bind the code delivery destination to the intended customer account.
That distinction is central to OTP authentication. A code establishes possession only when it reaches a destination already associated with the user. OWASP's testing guidance explicitly asks whether OTPs are tied to the correct user account. Reflectiz said that someone with the required customer identifiers could potentially exploit the flaw to access that customer's account.
From reconnaissance to validation
The agent first examined the site's JavaScript and the API calls supporting its pages. That reconnaissance exposed the OTP request and two values that appeared to control the delivery method and destination. Their presence alone was not proof of a weakness: the server could have rejected manipulated values or blocked the session at a later stage.
Reflectiz said the agent created a focused test around those values and executed it through the live login workflow. The successful authenticated state confirmed the bypass. Its guardrails then halted the test rather than allowing further activity in the account.
The company attributes this type of assessment to web context built while scanning thousands of production websites over more than a decade. For sites it already monitors, the agent can begin with established access and workflow knowledge rather than a blank URL. Reflectiz said its system drives a real browser through customer-facing, authenticated and multi-step processes, including OTP and two-factor authentication steps.
Coverage, repeatability and review
Reflectiz stressed that a single confirmed vulnerability is not a coverage report. It said the OTP test ran alongside other attack plans, while its structured process tests discovered endpoints against applicable attack categories. A separate validator agent reproduces each suspected finding using the same payload and context before a result is reported.
The insurer said it had pentested the site about six months before this assessment, although Reflectiz said it was unclear when the weakness had been introduced. The example illustrates the limits of a point-in-time test rather than a conclusion about human testers: a skilled tester could also have identified the flaw.
For businesses, the immediate control is to ensure that the server, not client-supplied request fields, binds every OTP channel and delivery address to the correct account. Teams should also retest login, recovery and other authenticated workflows after releases, with defined scope and guardrails and evidence that reviewers can verify.