Google Gemini API Keys Exposed in Android Apps
Researchers discovered that Google API keys embedded in over 500 million Android apps can gain elevated privileges when Google's Gemini API is integrated, potentially allowing unauthorized access to sensitive user data and services. Dozens of exposed API keys were found across apps including OYO, Google Pay for Business, Taobao, and others, with attackers able to extract keys through reverse-engineering.
What happened
CloudSEK researchers discovered that Google API keys embedded in Android apps, which were previously considered safe for client-side use, can gain elevated privileges once Google's Gemini Generative Language API is integrated. Analysis of 10,000 widely used Android apps found dozens of exposed API keys across apps with a combined install base of over 500 million. If a developer's API key is extracted by reverse-engineering the app, an attacker could gain access to Gemini endpoints without additional authentication.
Reported impact
Data shared with Gemini-powered features, including files, images and contextual AI interactions, can become accessible if keys are compromised. Developers face financial and regulatory risks as attackers can misuse these keys to make unauthorised API calls. Affected apps included OYO Hotel Booking App, Google Pay for Business (50M+ installs), Taobao (50M+ installs), apna Job Search App (50M+ installs), and ELSA Speak: AI English Learning (10M+ installs).
Index analysis
The vulnerability stems from API keys initially designed to function as identifiers for services like Maps or Firebase that can inadvertently become active credentials after Gemini's Generative Language API is enabled. According to KushoAI research, 34% of API test failures have direct security implications, with 38% of security failures attributed to authentication and authorization issues.
Governance controls
CloudSEK urged developers to audit their API key usage, avoid embedding sensitive keys directly in app code, and apply strict access restrictions. The report noted that current security testing practices have no coverage of supply chain risks targeting AI API credentials.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.