AI Agent Discovers Critical OTP Bypass on Live Insurer Website
An AI penetration testing agent autonomously identified and demonstrated a critical authentication bypass vulnerability on a major insurer's live production website, highlighting AI's growing offensive security capabilities.
An advanced AI penetration testing agent has autonomously discovered a critical authentication bypass vulnerability on a major insurer's live production website during an authorized assessment. The AI, developed by Reflectiz, successfully identified and exploited a flaw that allowed it to bypass One-Time Passcode (OTP) verification by manipulating the destination of the verification code.
The vulnerability stemmed from a flaw in the login process where a user could specify the delivery channel and address for the OTP. Instead of sending the code to the registered email or phone number associated with the account, the system would send it to an email address provided in the login request. The AI agent exploited this by directing the OTP to an inbox it controlled, thereby bypassing the intended security measure. This bypass was possible because the server did not validate that the provided email address was actually linked to the user's account.
Reflectiz's AI agent began its assessment by analyzing the website's JavaScript and API calls, which revealed fields that appeared to control OTP delivery. While these fields suggested a potential vulnerability, the agent proceeded to construct a targeted test case. By executing this test through the live login flow, the AI was able to confirm the bypass, demonstrating a significant capability in identifying complex vulnerabilities without human intervention.
The company emphasizes that the AI's ability to perform this test on a live production site was enabled by over a decade of prior work in scanning and interacting with thousands of production websites. This accumulated "web context" allows the AI agent to navigate multi-step workflows, including those involving OTPs and two-factor authentication, which often pose challenges for traditional automated testing tools.
This discovery underscores the evolving landscape of cybersecurity, where AI is increasingly being used for both offensive and defensive purposes. While the AI successfully identified the flaw, its guardrails were programmed to halt testing after the initial successful bypass, preventing further exploitation and ensuring ethical assessment practices.
The insurer, which reportedly had undergone a manual penetration test approximately six months prior, was alerted to the vulnerability. This incident highlights the limitations of point-in-time security testing against rapidly evolving threats and the potential for AI-driven tools to uncover vulnerabilities that might be missed by traditional methods.
Reflectiz positions its agentic pentesting as a more dynamic and comprehensive approach, capable of keeping pace with the rapid changes in web applications. The company claims its approach can offer up to 10 times the testing capacity at a similar operational cost to traditional manual pentesting, focusing on testing volume and depth rather than just finding more vulnerabilities.
The incident serves as a stark reminder that even seemingly simple flaws, when exploitable in production environments, can have critical security implications. The ability of AI to not only find but also demonstrate these vulnerabilities in real-world scenarios presents both opportunities for enhanced security and challenges for defenders.