Thousands of Google Cloud API keys available online may have given unauthorised access to sensitive Gemini AI endpoints, cybersecurity experts found. Security experts at Truffle Security discovered that around 3,000 live Google API keys were embedded in public websites, which could potentially be used for unauthorised access to sensitive data and billing abuse.
The problem occurred when users activated the Gemini API, or Generative Language API, under their Google Cloud projects.
Google Cloud API keys are used for project identification for billing and basic service access. However, Truffle Security discovered that when the Gemini API is activated, existing API keys in the same project automatically gain access to Gemini endpoints. Many of these keys were publicly accessible in website JavaScript code.
Truffle Security Researcher Joe Leon stated that attackers could access uploaded files and cached contents and create AI requests that could charge costs to the victims' accounts. The security firm found 2,863 live keys that could be accessed online, including one that belonged to a website associated with Google.
The keys could allow attackers to scrape websites, access endpoints such as /files and /cachedContents, and rack up large charges. In one case, reported on Reddit, a person stated that a stolen Google Cloud API key accumulated more than $82,000 in charges within two days.
Another piece of research carried out by a mobile security firm, Quokka, found more than 35,000 Google API keys embedded in Android apps. The research highlighted more worrying issues regarding AI endpoint abuse.
The issue has been acknowledged by Google, as the search engine giant’s spokesperson has said that the company has worked alongside researchers to resolve the problem, having taken measures to detect and prevent the leak of API keys that are attempting to access the Gemini network. It is, however, unclear if the vulnerability was exploited.