- A security researcher and their partner spent three months using an AI testing system to hunt for Google bugs, and ended up taking home $500,000 in bounties.
- They gathered about 3,600 API keys by scraping apps and intercepting network traffic across over 2,800 different Google domains.
- One critical bug allowed full account takeover on Google Voice and Fiber with just a single unauthenticated command.
A security researcher named brutecat turned an AI loose on Google’s infrastructure. The outcome? In less than 90 days, they received bug bounties of over $500,000 in total for finding security vulnerabilities.
The researcher’s campaign was able to uncover broken access controls across 1,500 different APIs. Many of these flaws sat hidden inside internal Google services.
The Hunt Started with API Keys
Brutecat began by targeting Google’s machine-readable API documents. These files list every available endpoint and method. Some are public, but many internal ones require valid API keys to see.
So the researcher teamed up with Michael Dalton to grab keys at scale. They scraped over 60,000 Android APKs. They also decrypted iOS binaries and built a Chrome extension to catch traffic across 2,800 Google domains. In the end, they collected about 3,600 keys.
A single key often unlocks multiple APIs on one Google Cloud project. This gave them wide access. They then filtered out non-Google keys using a Cloud Marketplace trick that identifies a project’s owner.
Uncovering Hidden Google APIs
The team found ways to see endpoints that Google tried to hide. They abused internal visibility labels like GOOGLE_INTERNAL. They also reverse-engineered Google’s own authentication system after the source code briefly leaked online.
In total, they gathered over 1,500 discovery documents from Google APIs. Brutecat then built a custom API Explorer that could parse any document and send authenticated requests.
Letting Claude AI do the Hacking
With the tools ready, the researcher added Claude AI as an engine for automated pentesting. The AI got three custom tools: confirm_testing_complete and probe_api, report__vulnerability. Its job was simple. Test every endpoint for broken access controls and IDOR flaws.
The team refined the system over a month. They taught the AI to group endpoints by type. Also, they made it probe each request across multiple API keys at once. They also had it translate Google’s cryptic error messages into plain English.
After these upgrades, the AI’s vulnerability reports were accurate more than half the time. Manual review became fast and easy.
A Very Dangerous Bug Found
The biggest discovery made was that the management API for Google Fiber and Google Voice, found at gfibervoice-pa.googleapis.com, does not have any access controls whatsoever. For an attacker, they would only need a single piece of information: a victim’s Gaia ID.
With the Gaia ID, the attacker could then issue a command via this API, without needing any authentication, and retrieve complete personal information about the victim, including the victim’s Google Voice phone number and account recovery phone number.
To make the situation worse, attacking this API allows an attacker to associate any phone number they want with the victim’s account. And that phone number would be listed in the victim’s verified list of phones.
Google rated this bug as a P0/S0 critical bug (the highest severity level possible). They patched the vulnerability within hours after being notified and paid the researcher who discovered this flaw $20,000 for their submission.
Total Amount of Bounty Paid for All Bugs
Some other notable bugs the researcher found included an AdExchange takeover and Google paid them $30,000 for it. Also, they uncovered a flaw in the Translation Hub and got paid $36,500.
The researcher also discovered another bug within the Cloud Console GraphQL service (CVE-2026-8934). That bug could leak App Engine request logs without requiring any type of authentication.
Overall, Google paid out more than $500,000 in Bug Bounty to the researcher for uncovering all these bugs. And the researcher reported all of them through the Google Vulnerability Reward Program.
Major tech companies regularly face security challenges. Facebook previously confirmed a bug that exposed up to 6.8 million users’ unposted photos, underscoring the importance of robust security testing.
This campaign shows a major shift in security testing. AI is no longer just a helper. In the right hands, it becomes a powerful discovery engine that can find critical flaws even inside the world’s most locked-down companies.
Share this article
About the Author
Farwa is an experienced InfoSec writer and cybersecurity journalist skilled in writing articles related to cybersecurity, AI, DevOps, Big Data, Cloud security, VPNs, IAM, and Cloud Computing. Also a contributor on Tripwire.com, Infosecurity Magazine, Security Boulevard, DevOps.com, and CPO Magazine.
More from Farwa SajjadRelated Posts
Google and FBI Disrupt NetNut Residential Proxy Network Used by 2M+ Devices
Google worked with the FBI and Lumen to disrupt the NetNut residential proxy network, also known as ...
Opera Launches Paste Protect to Block ClickFix Clipboard Attacks
Opera’s Paste Protect blocks harmful commands from reaching the clipboard, preventing ClickFix...
Reddit Introduces Mandatory Age Verification for EU Teens Accessing NSFW Content
Reddit will require European Union users under 18 to verify their age before viewing mature or NSFW ...
Popular ‘Adblock for YouTube’ Chrome Extension Found With Remote Code Execution Risk
The popular “Adblock for YouTube” Chrome extension now carries an architectural weakness...
Texas Hunting and Fishing License Data Breach Affects 3 Million Customers
Approximately 3 million Texas hunting and fishing license customers were affected by a data breach i...
FBI Warns of ‘Kali365’ Subscription Service Targeting Microsoft 365 Accounts
The FBI’s sounding the alarm on Kali365, a site where criminals can pick up ready-to-use tools for s...
