Two of the Google Cloud developers who were hit with bills for thousands of dollars following unauthorized API calls to Gemini models have had their bills reversed, the users told The Register in recent days. But Google plans to continue automatically expanding users’ spending limits, leaving them and countless other customers vulnerable to bills they cannot afford, whether from fraud or a sudden traffic surge.
Australia-based developer Isuru Fonseka – whose usage bill skyrocketed to $17,000 in minutes after Google automatically upgraded his $250 spending tier when a hacker took control of his account – told us that he was happy to put this behind him.
“It’s so good. It felt like they were just giving me the run around until your article. I just hope they fix it properly for everyone,” he said. “It’s great that the article was able to get the refund but it’s sad that it had to go to that level for them to process it urgently.”
Despite refunding his money, Google seems to have lost a customer. Fonseka said that he has since ensured his API cannot be used with Google’s stable of AI products, and will likely try one of the independent foundation models if he needs those features.
“I’ve disabled Gemini on everything – if I ever plan to use AI on my projects, I’m better off using it via a different service such as OpenRouter or going directly to one of the other LLM providers – just as a way to keep Gemini out of my account and the risk as low as possible,” he said.
Fonseka said he was blindsided by a Google policy that allowed the company to automatically upgrade a user’s billing tier without permission or adequate warning. He had thought by signing up for a user tier with a $250 spending cap that his bills would be restricted to that amount. It was only after attackers exploited his API key that he learned Google would upgrade the cap automatically based on his history of spending.
While Google acknowledged that the automatic tier upgrades allowed credential hijackers to rack up thousands of dollars in bills in cases like the one Fonseka described to The Register, it said it has not reconsidered the policy.
In a statement to The Register, Google said that it wants to prioritize access to Google Cloud services without interruption, preferring to prevent service outages over respecting users’ budget preferences.
“With our automated growth tiers, we helped businesses scale as usage increased, built on their historic reputation of payments and usage,” a Google spokesperson told us in a statement. “This prevents their business having a hard service outage once they pass an artificial system quota.”
Tiers vs spending caps
There is some confusion between Google’s usage tiers and its newly introduced spending caps, and Google’s documentation hasn’t helped much.
Google says its users can set their usage tiers not to exceed a certain spending level. For example the maximum spending allowed by a Tier 1 user like Fonseka is $250. However, if the account is older than 30 days and if, over the lifetime of their work with Google, they have spent at least $1,000, then Google will automatically allow that account to spend up to $100,000. So good customers have the most to fear from fraud or from an unexpected spike in usage.
In several cases shared on social media, Google users were only aware of this after their credit cards were billed thousands of dollars.
On April 22, Google introduced a trial of hard caps on spending within Google Cloud, but those are in a preview and are approved on a case-by-case basis.
“We’re excited to announce that Spend Caps are coming soon to Google Cloud. Designed to work with Google Cloud Budgets, FinOps and DevOps can set budgets that enforce automated cost boundaries (caps) at the project level for AIS, Agent Platform, Cloud Run, Cloud Run Functions, and Maps,” Google wrote. “These caps alert and ultimately pause API traffic once your set budget is reached, but leave your resources intact. If you need the traffic to resume, simply suspend the Spend Cap.”
Spend caps can only be set per project for a single, eligible service, Google said. Eligible services for this preview include Gemini API, Agent Platform (previously known as VertexAI), Cloud Run, Cloud Run Functions, Maps, Google said.
Users who apply for a spending cap will have their submissions reviewed on a “one to two week basis” and customers are added in the order they submitted.
“Once onboarded, you will receive an email with instructions on how to access the feature as well as details on how to submit feedback,” Google writes in its sign up page.
Rod Danan, CEO of Prentus, a company that helps job applicants with interview preparation and tracks job placements for universities, told The Register earlier this week that he saw his bill skyrocket to $10,000 in just 30 minutes of usage by attackers who exploited his public API key.
Google forgave the charges on Thursday, he said.
“They got back to me today agreeing to a refund,” he told us. “It’s definitely relieving. You want to focus on the business. You don’t want to have to focus on going and getting refunds from some crazy charges.”
He said the stress of running a startup is hard enough without the addition of fighting one of the largest companies in the world imposing erroneous five-figure charges.
“I’m happy that it’s behind me. I wish it was easier,” he said. “I’ve learned, yeah, definitely don’t give up. Be annoying whenever something is wrong and just keep pushing. Again, try to make it as public as possible, get louder and louder until the people you need to hear you actually hear you.”
Google said any unauthorized use of API keys will be investigated and it historically has treated customers compassionately when there is clear evidence of fraud or error.
“We take reports of credential abuse and the financial security of our customers extremely seriously; and as you know are investigating these specific cases you have pointed to and we will work directly with any impacted users to resolve charges resulting from fraudulent activity,” Google said. ®
