AN AI-assisted pentest uncovered a critical authentication bypass on a major insurer’s production website during an authorised assessment. The Reflectiz AI agent mapped the login flow and attempted a real login by directing the one-time passcode (OTP) to an email inbox it controlled. The login succeeded, but the test was halted by the defender’s guardrails after the login completed, revealing a serious flaw rather than a mere probe. The insurer was informed of the finding.
The flaw stemmed from how the OTP was delivered. The login request included two fields that dictated delivery: the channel and the address. Ordinarily the OTP would be sent to the registered phone number, but the request allowed changing the channel to email and specifying any address in the payload.
In effect, the server did not verify that the chosen address belonged to the account, meaning an attacker with the required customer identifiers could potentially receive the OTP and gain access to the customer’s account. Binding an OTP to an account is a fundamental authentication check, and the omission here enabled an authentication bypass in a production environment.
The agent progressed from initial reconnaissance of site scripts and API calls to a targeted live test, confirming the bypass with a controlled test flow. Reflectiz reported the issue to the insurer. The article emphasises that web context behind the login gate—gaining access to authenticated workflows, verification steps, and multi-step processes—was essential to proving the bypass, something the agent’s long-built context helped achieve. The insurer’s response and any remediation steps are not detailed in the report.