Multiple developers have received bills running into tens of thousands of dollars for Gemini AI API usage they never initiated. The root cause is tangled API management inside Google Cloud. Any API key created through the platform could call AI services and more, even if the developer originally created that key solely for Google Maps.
How Old Keys Became a Liability
The problem traces back to poor permission isolation. Years ago, a developer might have requested a key to call Google Maps inside an open-source application. That call was free, so publishing the key in a configuration file carried no apparent risk.
The landscape has since changed. Attackers now scan open-source repositories for valuable keys. Any Google Cloud API key qualifies, because those same keys can also invoke the Gemini AI API. Developers who never used Gemini thus found themselves unknowingly billed for thousands of dollars of AI usage driven by stolen credentials.
Google has waived the charges in many cases. However, not every developer has been so fortunate. Some had to post publicly on social media before the Gemini team took notice. Only now is Google preparing to close the loophole for good.
Standard Keys Lose AI Access in September
On 30 July 2026, Google updated its developer documentation for the Gemini API. The change is direct. To ensure secure usage, Gemini will switch from standard keys to authenticated keys. Starting in September 2026, all requests made with standard keys will be blocked. Developers must migrate to the new authenticated keys before that deadline to avoid disruption.
The distinction between the two key types matters. A standard key ties to a Google Cloud project but cannot identify the caller. It therefore cannot deliver the fine-grained permission and access controls the situation demands.
An authenticated key, by contrast, binds directly to a Google Cloud service account. It provides granular access control, and by default it works only with AI APIs.
What Developers Need to Do
After September, any developer calling Gemini must hold an account-level authenticated key created specifically for AI services. Project-level standard keys will no longer work for AI calls.
Developers can edit the permissions on an authenticated key to grant access to additional services if needed. For security, though, Google recommends using separate keys across different projects to keep controls tight and segmented.
Key management remains available through Google AI Studio. Developers should consult the updated support documentation for full migration steps.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
