Two Google Cloud customers said in May 2026 that unauthorized API use had generated thousands of dollars in AI charges, despite spending limits they believed would contain their bills. Their accounts concern access to costly image and video generation services and usage-tier ceilings that Google says can increase automatically.
Google said compromised credentials were an industry-wide problem rather than a security issue specific to its services. The company said most such incidents involved exposed credentials, including API keys leaked in public code repositories.
Customers describe sudden billing spikes
Rod Danan, CEO of Prentus, said his company used Google Maps API calls in its product and had paid no more than $50 a month for years. In March, he said, he received a $3,000 billing alert, followed by another $5,000 charge about five minutes later while he tried to identify the usage.
Danan put the total at $10,138 by the time he disabled the API. He said nearly all the charges involved Veo 3 video generation and Gemini image output tokens, services his product did not use. He also said Google had told him it found no evidence of fraud and had refused a refund.
Sydney developer Isuru Fonseka said he received alerts on April 29 after using Google Cloud for 10 years. He put the charges at roughly AUD $17,000 and said his card provider declined some attempted transactions. Fonseka said he had never exposed the API key he used with Firebase.
Fonseka said he had expected a $250 Tier 1 limit to constrain spending but later saw a tier ceiling of about $100,000.
Automatic tier increases and separate project caps
Google said it believed attacker-driven usage had triggered an automatic tier upgrade in Fonseka’s case, rather than an attacker changing the tier manually. The company cited Tier 3 qualification criteria of at least US$1,000 in Cloud payments and 30 days since the first payment.
In its March 16 announcement about Gemini API costs, Google described automatic upgrades as usage and payment history met the criteria for higher tiers. The company said each tier had a system-defined monthly spending ceiling across the billing account, which increased as the account advanced.
Google distinguished those tier ceilings from custom Project Spend Caps. Its description of project caps said user-configured limits remained active until changed or disabled and operated independently of the automatically increasing tier ceilings. Google also noted an approximately 10-minute enforcement delay and said customers remained responsible for overages incurred during that period.
Google’s billing guidance separately described alerts-only budgets as tools for monitoring costs and triggering notifications, distinct from spend-cap controls.
Public Maps keys and Gemini access
Separate research published in February examined how publicly visible Google API keys could also provide access to Gemini. Truffle Security said a scan of the November 2025 Common Crawl dataset identified 2,863 live Google API keys exposed on websites that could authenticate to Gemini.
Joe Leon, the Truffle Security researcher behind the findings, said enabling the Gemini API in a Google Cloud project could give existing keys in that project access to Gemini. The concern included keys deployed publicly for Maps whose restrictions did not prevent access to the AI service.
Google said it had made API restrictions mandatory when creating keys and that it was no longer possible to create a key with access to both Maps and Gemini. The company also advised against using the same key across multiple APIs, particularly when visible in a browser, and recommended restrictions on both permitted services and the applications allowed to use a key.
