NOTE: As of the publication of this article, the vulnerability has been fully patched, and all coordination regarding disclosure was managed directly with the Google VRP team.

Introduction: “Security Architecture vs. The Real World”

How can Android’s foundational security layer, the Keyguard, falter when confronted with the complexity of a modern AI interface? The answer is simple: as systems grow more complex, the impact of overlooked edge cases amplifies.

In this write-up, I will dissect how a simple multi-touch interaction triggered a critical logical security flaw. I’ll provide the technical details of how this vulnerability bypassed the lock screen — the very boundary designed to be the most secure — and exposed sensitive user data.

How I Discovered It

I didn’t uncover this vulnerability using automated scanning utilities or complex fuzzing frameworks; rather, it surfaced organically during my daily user experience. I noticed that multi-finger interactions within the user interface triggered benign functions that should have been restricted under specific device states. While an average user might dismiss this behavior as a transient UI glitch, to a security researcher, it signaled a potential flaw.

Further analysis revealed that invoking specific operational modes, such as Lyria or Deep Research, forced the application into a full-screen state. Initially, interaction was restricted, and the system prompted for credentials. However, applying the multi-touch technique I discovered earlier circumvented these constraints, granting unauthorized access to application settings and chat histories. By expanding the attack surface during my research, I confirmed that critical assets like NotebookLM notebooks and Gmail drafts were equally vulnerable, subsequently documenting and escalating these findings to Google.

Technical Analysis

Gemini’s modular features bypassed the system Keyguard due to an architectural misconfiguration. The application improperly exposed UI elements that should remain strictly inaccessible while the device is locked. Although the system repeatedly invoked the Keyguard to request authentication during these operations, I successfully bypassed the lock state by leveraging a form of Context Hijacking.

The root cause lies in inadequate validation of concurrent UI interactions. By maintaining an active press on a permitted interaction area (such as a text input field) while simultaneously tapping a restricted target element, the application failed to isolate the input contexts. This race-like UI interaction completely neutralized the application’s internal security control mechanisms, turning a seemingly minor interface bug into a robust logical bypass.

Impact & Exploitation Surface

During the initial phase of my research, the exploit vectors were limited to reading, deleting, or renaming historical chats, accessing Gemini’s core settings, and viewing profile data. However, digging deeper into the application’s ecosystem revealed a significantly more severe impact:

• Arbitrary Creation of Gmail and Google Docs Drafts: Allowing unauthenticated data injection into core Google services.
• Unauthorized Access to NotebookLM: Exposing proprietary or highly sensitive personal and enterprise data stored within notebooks.
• Gem Execution: Triggering custom AI agents without owner authentication.
• Destructive Actions: Permanently deleting critical NotebookLM assets.

The practical implications of this vulnerability present severe risks, including data exfiltration, advanced social engineering scenarios via unauthorized draft creation, and the compromise of enterprise-grade environments.

Coordination and Disclosure Timeline

Throughout the lifecycle of this vulnerability, I maintained an active and transparent line of communication with Google’s security team. Shortly after submission, my report was designated as a “Duplicate,” tied to an older legacy issue inherent to Android’s core component architecture. Despite my requests for verification regarding the unique interaction vector, the root cause was maintained as identical.

Nevertheless, I continued my research. Following a subsequent major Gemini update, I verified that the exploit remained active. Upon presenting this evidence, an immediate mitigation was deployed, removing the specific mode buttons from the locked interface. Roughly a month later, a comprehensive patch was pushed, completely resolving the underlying logic flaw. My subsequent regression testing confirmed that unauthorized multi-touch access had been completely mitigated, successfully concluding the lifecycle of the report.

Key Takeaways and Conclusion

This journey marked my very first experience within the bug bounty ecosystem. Uncovering this logical vulnerability taught me that security research extends far beyond hunting for code flaws; it is about navigating the disclosure process and understanding how even a “duplicate” report can be leveraged to harden a system’s overall security posture. It perfectly illustrated how seemingly decoupled, non-critical components can be chained together to form a high-severity exploit.

Analyzing Android’s security architecture and engaging with engineering teams on regression analysis during my first research attempt fundamentally shifted my perspective on information security. While this specific report did not yield a financial bounty, the true payout was invaluable: a deep dive into the inner workings of complex enterprise software and the discipline required to execute a responsible disclosure process.

This experience is merely the opening chapter of my career in security research. It stands as a reminder that no architecture is infallible, but through the vigilance of independent researchers, they can be made resilient. Security is never a static defense; it is a continuous cycle of curiosity, analysis, and refinement.

NOTE: All PoC media provided in this article have been redacted to ensure user privacy and are presented solely for educational and security analysis purposes.

📌 References & Community

If you want to check out my other security research, tools, or open-source projects, feel free to explore the links below:

• GitHub: github.com/msalihberk
• Previous Research: ShadowLab: A Modular C2 Framework Architecture Built with Python for Modern Cybersecurity Research
• Follow for More: Feel free to follow my Medium profile to get notified about my future security research, development projects, and technical write-ups.