The vulnerability reflects a structural issue: identifiers that were once harmless in older development practices effectively became authentication credentials without clear safeguards.

Keys once seen as harmless public identifiers are now sensitive AI credentials, elevating the impact of any exposure.

In decompiled apps, live Gemini credentials can be extracted, enabling unauthorized access to Gemini-shared data such as documents, images, audio, and cached AI context.

Attackers can generate large volumes of AI requests, leading to significant unintended charges and quota exhaustion for developers.

A security finding by CloudSEK reveals 32 Google API keys embedded in 22 popular Android apps, collectively serving over 500 million users, which can grant unauthorized Gemini access when the associated project’s AI is enabled.

These keys follow the standard API key format (AIza…) and can enable retroactive privilege escalation, allowing attackers to issue Gemini API calls, exhaust quotas, and access data stored in Gemini, including documents and images.

The finding highlights how legacy development practices clash with rapid AI infrastructure adoption, creating security gaps at scale.

Gemini integration costs fall on developers, and data breaches could trigger regulatory scrutiny for the affected companies.

Financial damages from similar exposures have included tens of thousands to over a hundred thousand dollars in a matter of hours for individual developers and companies across different regions.

Exposure spans multiple sectors, including travel, fintech, and education, with apps like OYO, Taobao, Google Pay for Business, and JioSphere affected.

The report emphasizes the need for improved API key management and safeguards to prevent misuse of embedded keys in mobile apps.

Keys can persist across app updates, creating long-term risk of ongoing unauthorized access until mitigated.