Major Google Bug Triggers Gemini AI Leak in Google Pay for Business, Oyo Hotels and Other Apps: Report

Google's implementation of an application programming interface (API) key architecture has reportedly led to a massive Gemini exposure risk in Android apps. As per the cybersecurity research firm CloudSEK, a particular client-side API key, which previously functioned as an identifier, receives credential privileges after an Android app integrates the Gemini API. This, in the hands of a bad actor, can expose the data users share with the chatbot. Additionally, this can also result in bad actors making unauthorised Gemini API calls, racking up huge bills for the developer.

How a Google API Key Triggers Gemini Exposure

In a blog post, CloudSEK explained how an API key (AIza...), which was deemed safe by Google to add to the codebase of Android apps, suddenly gains credential privileges after Gemini is integrated into the app. This security flaw builds on the findings of Truffle Security, which found a similar flaw on a Google Cloud project.

CloudSEK's BeVigil, a mobile app security search engine, scanned the top 10,000 Android apps (based on number of installs) and found 32 live Google API keys hardcoded in 22 different apps with more than 500 million installs collectively. Some of these apps are Oyo Hotel, Google Pay for Business, Taobao, apna Job Search App, Elsa Speak, HD Sticker & Pack WAStickersApps, The Hindu, ISS Live Now, and more.

Interestingly, the report claims that the API key format Alza… is added to the app when a developer wants to embed Maps or Firebase, as per the documentation instructions shared by Google. However, after enabling the Generative Language API, the key gains access to all Gemini endpoints without any warning or notification. So, anyone who decompiles the app can easily gain access to the key, and it acts like a live Gemini credential.

For end users, this means any data shared with Gemini, such as documents, images, or audio, and stored in the Files API, can be accessed by the bad actor. Additionally, all sensitive information in the cached AI context can be read, copied, or exfiltrated by the one gaining unauthorised access.

Developers and publishers also face significant risks. Gemini API integration is not free. Developers pay for any usage. So, if the bad actor ends up making unauthorised usage, it can rack up massive bills. Additionally, this also puts a regulatory burden on the companies if the users' data is compromised.

CloudSEK recommends developers and companies review all API keys in a GCP project, rotate any key that is embedded in a mobile app, restrict keys by service, and not hardcode any API key in the mobile app source code. While end users cannot do much, they should be careful about using Gemini services in an Android app. If they do not trust the app, they should limit their Gemini interaction to the official app and platforms.