AUD $50,313 unauthorized Gemini API usage: $12,626 remaining after five-month investigation

Google Cloud Support case: 70166119

I’m seeking help from a Gemini/Google AI staff member to route an unresolved billing dispute for further review. Approximately AUD $50,313 in unauthorized Gemini API usage occurred on April 13, 2026. Google has adjusted 75% of it, but Cloud Billing has now closed the case leaving AUD $12,626.20 payable and has declined further escalation.

What happened

  • The affected project had effectively zero usage immediately before and after the incident, and I had not personally used the affected credential for approximately a month.
  • On April 13, approximately AUD $50,313 of Gemini API usage occurred in an extraordinary isolated spike and then returned to effectively zero.
  • I reported the incident to Google and immediately disabled billing on the affected project and revoked/rotated relevant credentials. I subsequently added further project isolation and spending controls.
  • After an investigation lasting more than five months, Google confirmed that the April 13 usage was unauthorized and credited AUD $37,686.84, representing 75% of the unauthorized charges.
  • The remaining AUD $12,626.20 is still being charged to me and remains formally disputed.

The unresolved issue

Google initially told me its engineering/security investigation determined that the unauthorized usage resulted from “a leaked API key on the user end.”

When I asked for the basis for that attribution, Google subsequently clarified that its investigation established that the unauthorized requests were authenticated using a valid API key associated with my project and that Google found no evidence of a Google system error or infrastructure breach.

I do not dispute Google’s finding that a valid API key authenticated the requests. An API key may legitimately have existed for AI Studio/local Gemini testing.

What remains unexplained is how an unauthorized third party obtained that credential.

To the best of my knowledge, the relevant credential was never committed to a public repository or embedded in a publicly deployed application, and no online project of mine was using an exposed Gemini API key. My production application’s normal Gemini authentication flow uses OAuth/service-account authentication rather than an API key.

Google has not identified an exposure attributable to my handling of the credential. Its final position is instead that because the unauthorized requests were authenticated using a valid project credential, the resulting charges are the responsibility of the billing account owner.

Why I am asking for another review

This incident occurred during a period in which other developers publicly reported sudden unauthorized Gemini API usage resulting in substantial unexpected charges.

The Register reported on cases involving Rod Danan and Australian developer Isuru Fonseka, both of whom experienced substantial unauthorized Gemini API charges despite reporting that they had spending controls in place. Google subsequently reversed their unauthorized charges.

I am not claiming that the compromise mechanism in those cases was identical to mine. I am pointing to them because they demonstrate that unauthorized use of a valid Google API credential has not invariably been treated as sufficient reason to leave the resulting charges with the customer.

Google has also since strengthened Gemini API credential security. It now rejects unrestricted standard keys, defaults new AI Studio keys to service-account-backed authorization keys, and is transitioning away from standard keys entirely. Google’s documentation describes the change as preventing unauthorized use of keys that may be shared publicly or linked to other services, while authorization keys provide fast-acting enforcement against leaked keys detected by Google’s systems.

I am not claiming that the specific security issues addressed by those changes caused my April 13 incident. I include this as relevant context given the broader credential-security problems occurring around the same period.

What I am asking

Could a Gemini/Google AI staff member please help route case 70166119 to a team capable of reviewing the remaining AUD $12,626.20?

The ordinary Cloud Billing process has now been exhausted. Support has told me that its 75% goodwill adjustment is the maximum allowable adjustment, that the remaining balance is my responsibility because a valid project credential was used, and that it will not escalate the case further.

I can provide the complete five-month support correspondence, billing information and other relevant details privately to Google staff if required.

Thank you.