Google API keys, once considered harmless when embedded in public websites for services like Maps or YouTube, have turned into a serious security risk following the integration of Google’s Gemini AI assistant. Security researchers at Truffle Security uncovered this issue, revealing that nearly 3,000 live API keys—prefixed with “AIza”—are exposed in client-side JavaScript code across popular sites.
Truffle Security’s scan of the November 2025 Common Crawl dataset, which captures snapshots of major websites, identified 2,863 active keys from diverse sectors including finance, security firms, and even Google’s own infrastructure. These keys, deployed sometimes years ago (one traced back to February 2023), were originally safe as mere billing identifiers but gained unauthorized access to Gemini endpoints without developers’ knowledge.Attackers can simply copy a key from page source, authenticate to Gemini, and extract sensitive data like uploaded files, cached contexts, or datasets via simple prompts.
The danger extends beyond data theft to massive financial abuse, as Gemini API calls consume tokens that rack up charges—potentially thousands of dollars daily per compromised account, depending on the model and context window. Truffle Security demonstrated this by querying the /models endpoint with exposed keys, confirming access to private Gemini features. One reported case highlighted an $82,314 bill from a stolen key, underscoring the real-world impact.
Google acknowledged the flaw as “single-service privilege escalation” after Truffle’s disclosure on November 21, 2025, and implemented fixes by January 2026, including blocking leaked keys from Gemini access, defaulting new AI Studio keys to Gemini-only scope, and sending proactive leak notifications. Despite these measures, the “retroactive privilege expansion” caught many off-guard, as enabling Gemini in projects silently empowered old keys.
Developers must immediately audit Google Cloud projects for Gemini API enablement, rotate all exposed keys, and restrict scopes to essentials—avoiding the default “unrestricted” setting. Tools like TruffleHog can scan code repositories for leaks, while regular monitoring prevents future exposures in an era where AI services amplify API risks. This incident highlights the need for vigilance as cloud features evolve.
This article has been indexed from CySecurity News – Latest Information Security and Hacking Incidents
Read the original article: