The story of Reflectiz’s AI agent finding a critical OTP bypass during an authorized pentest of an insurer’s production site is less about the vulnerability itself and more about what the agent did not do next.
The flaw was straightforward: send an OTP to any email address you control, complete the login. A human pentester would have done exactly that. So did the agent. Then it stopped.
Agentic pentesting can find flaws at scale that manual testing misses, but the real vulnerability is not in websites. It’s in deployment without restraint. Reflectiz’s agent had guardrails. The ARTEX agents hitting South Korean banks in September had different marching orders.
This case is reassuring as proof that the tool works, the guardrails worked, and responsible disclosure happened. But it also shows a thin margin: the gap between pentesting breakthrough and account takeover at scale. That margin depends entirely on whoever’s holding the keys.
Sources: An AI Agent Found a Critical OTP Bypass in Production • Reflectiz Launches Agentic Pentesting for Websites
Coverage: South Korea Bank Hacks 2026: ARTEX and Agentic Pentesting
Related on this blog: [AUTO] When your AI agent decides unauthorized access is a reasonable tactic • [AUTO] AI-Orchestrated Campaign Compromised 395 Organizations • [AUTO] Workflow Identity Hijacking: The Authorization Gap