During an authorized assessment, Reflectiz's AI pentesting agent was mapping the login flow of a major insurer's production website, which serves millions of users. The agent requested a login code and directed it to an email inbox it controlled. The code arrived, the login succeeded, and the agent's guardrails halted the test. The agent had independently discovered a critical-severity authentication bypass.
The flaw was simple: the login request carried two fields that decided how and where the one-time passcode was delivered. Normally, the code went by text to the phone number on the account. But by changing the channel to email and specifying an arbitrary address, the server sent the code to that address without ever checking whether it belonged to the account. An attacker with the required customer identifiers could potentially use this to access the customer's account.
Proving the flaw on a live site was the hard part, according to the source. The agent's analysis of JavaScript and API calls surfaced the suspicious fields, but that only suggested a hypothesis. The agent then built a targeted test and ran it through the live login flow, confirming the bypass. Reflectiz attributes this capability to more than a decade of scanning production websites, which gave the agent context for navigating authenticated, multi-step workflows.
The insurer had reportedly pentested the site about six months earlier, but the weakness was still present. The article argues this reflects the limits of point-in-time testing: websites change with every release, while pentest schedules are rigid. Agentic testing can repeat focused tests as often as the site changes, and Reflectiz reports up to 10x the testing capacity at the same operational cost as manual pentesting—though the article clarifies this is a volume comparison, not a guarantee of finding 10x more vulnerabilities.