Researcher Hacks Google Using AI, Earns $500,000 Bug Bounty

Key Takeaways

• A security researcher leveraged an AI-powered fuzzing pipeline to discover over $500,000 worth of vulnerabilities within Google’s extensive infrastructure in less than three months.
• The AI system identified systemic access-control issues across approximately 1,500 Google APIs.
• One critical finding, rated P0/S0, allowed unauthenticated access to Google Voice and Google Fiber user PII and enabled arbitrary phone number assignment, posing a significant account takeover risk.
• Google has patched the identified vulnerabilities, demonstrating the effectiveness of AI in uncovering severe flaws in large-scale systems.

AI-Driven Fuzzing Uncovers Half-Million Dollar Flaws in Google APIs

In a groundbreaking demonstration of artificial intelligence’s potential in offensive security, a researcher known as brutecat has revealed how an AI-powered fuzzing pipeline successfully identified over $500,000 in vulnerabilities across Google’s vast digital ecosystem. Over a period of less than three months, this sophisticated system systematically exposed widespread access-control deficiencies embedded within approximately 1,500 distinct APIs.

Methodology: From Discovery Docs to API Key Harvesting

The research initiative commenced by targeting Google’s machine-readable API specifications, often referred to as discovery documents. Similar to Swagger documentation, these files enumerate all available API endpoints, parameters, and methods. While public discovery documents exist for services like the YouTube Data API, a substantial number pertain to internal Google APIs, necessitating valid API keys for access.

To overcome this hurdle, brutecat collaborated with Michael Dalton to collect credentials at scale. Their strategy involved scraping over 60,000 Android APKs, decrypting iOS binaries, and developing a Chrome extension to intercept network traffic across more than 2,800 Google web domains. This extensive effort ultimately yielded approximately 3,600 API keys.

Given that a single API key can often enable multiple APIs within its associated Google Cloud project, this collection of credentials provided broad access. To ensure compliance with Google’s bug bounty program scope, the team filtered out non-Google keys by utilizing a Cloud Marketplace endpoint capable of resolving a project number to its owning domain.

Bypassing Protections and Building an AI Pentesting Engine

The researchers further advanced their access by bypassing previously removed discovery paths. They exploited visibility labels, such as GOOGLE_INTERNAL, to expose hidden endpoints and successfully reverse-engineered Google’s proprietary First Party Authentication (FPA v2) after relevant frontend library sourcemaps were briefly leaked.

Having amassed over 1,500 discovery documents from Google APIs, including those with hidden endpoints revealed by undocumented GOOGLE_INTERNAL labels, brutecat developed a custom API Explorer. This tool was designed to parse any discovery document client-side and execute authenticated requests.

With this robust infrastructure in place, the researcher integrated Claude AI to function as an automated penetration testing engine. The AI was equipped with a specialized toolkit, including functions like probe_api, report_vulnerability, and confirm_testing_complete, enabling it to systematically scrutinize every endpoint for broken access controls and Insecure Direct Object Reference (IDOR) vulnerabilities.

The AI system underwent a month-long refinement process through iterative prompt engineering. Key enhancements included group-based endpoint classification, multi-key probing (which automatically dispatched the same request across all known API keys), and standardized parsing of Google’s often cryptic API error messages into more comprehensible labels. Following these improvements, the AI’s vulnerability reporting accuracy surpassed 50%, significantly streamlining the manual review process.

Critical Findings and Bounty Payouts

Among the most severe vulnerabilities uncovered was a complete absence of access controls on gfibervoice-pa.googleapis.com, an API responsible for managing Google Voice and Google Fiber accounts. This critical flaw, rated P0/S0 by Google, allowed an attacker to retrieve full Personally Identifiable Information (PII), including a victim’s Google Voice number and account recovery phone number, using only an unauthenticated curl command and the victim’s Gaia ID.

Even more alarmingly, the API permitted an attacker to assign any phone number to a victim’s Google account without authorization. The newly assigned number would then appear under the victim’s verified phones at myaccount.google.com/phone. This vulnerability presented a clear pathway for potential account takeovers (ATO) and SIM-swap-style attacks. Google promptly patched this bug within hours and awarded $20,000 for this single discovery.

Other significant vulnerabilities identified and their respective bounties include:

All identified vulnerabilities were reported responsibly through Google’s Vulnerability Reward Program (VRP). The AI-assisted research campaign ultimately uncovered bugs across dozens of internal Google APIs, resulting in collective bounty payouts of $500,000 to the researcher in under 90 days.

This research highlights a significant evolution in offensive security, demonstrating that AI is no longer exclusively a defensive tool. In skilled hands, it transforms into a highly scalable vulnerability discovery engine, capable of unearthing critical flaws even within the most security-conscious organizations globally.

What You Should Do

• Organizations: Consider integrating AI-powered fuzzing and security testing into your development lifecycle to proactively identify complex vulnerabilities, especially in extensive API ecosystems.
• API Developers: Implement robust access control mechanisms at every layer of your API architecture and conduct thorough security audits to prevent issues like IDOR and unauthenticated data access.
• Security Teams: Stay informed about advanced AI-driven security testing methodologies and evaluate their applicability for enhancing your own vulnerability discovery processes.
• Users: While this research concerns Google’s infrastructure, it underscores the importance of strong account security. Enable multi-factor authentication (MFA) on all your accounts, especially those with sensitive information, and be vigilant against phishing attempts that could compromise API keys or other credentials.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.