Google Cloud’s security advice clashes with developer billing reality

Francis deSouza, the COO of Google Cloud, recently asserted that security cannot be an afterthought for companies using artificial intelligence. Speaking on the industry’s shift toward AI adoption, deSouza stated, “There’ll be a transition period, and then I think we get to this better place.” He warned that there is no AI strategy without a data strategy and a security strategy, urging organizations to avoid “shadow AI”—which refers to employees using consumer AI tools without organizational oversight. To counter rapidly evolving threats, deSouza noted that the average time between an initial breach and the handoff to the next stage of an attack has dropped from eight hours to 22 seconds. He advocated for “agentic defense,” or AI-native security systems using agents to defend infrastructure, arguing, “As companies embark on this AI journey, they need to take a platform approach”.

Despite these high-level warnings, Google Cloud developers hit with five-figure bills following unauthorized API calls to Gemini models have raised concerns about platform-level safeguards. These incidents occurred after attackers exploited compromised API keys. Google has stated it has no plans to change its automatic tier-upgrade policy, which can raise billing ceilings as high as $100,000 based on account history.

Specific billing incidents include:

• Rod Danan, the CEO of Prentus, saw his bill reach $10,138 in roughly 30 minutes after attackers exploited his compromised API key.
• Isuru Fonseka, a developer, was charged roughly AUD $17,000 despite believing he had set a $250 spending cap.

The vulnerability is compounded by delays in credential revocation. Security firm Aikido reported that attackers can apparently continue using a deleted API key for up to 23 minutes because Google’s revocation propagates gradually. During this window, Aikido researcher Joseph Leon found that over 90% of requests still authenticated in some minutes, allowing attackers to continue exploiting the keys. This delay contrasts with Google’s newer credential formats: service account API credentials revoke in five seconds, and newer AQ-prefixed keys take about a minute. Addressing the broader landscape of vulnerabilities that AI itself is introducing, Lea Kissner, LinkedIn’s chief information security officer, noted, “We’re going to need people to deal with the bug-pocalypse”—referring to a massive influx of software vulnerabilities.

Why it matters

This situation highlights a critical gap between the high-level security advice provided by major cloud platforms and the actual operational risks faced by developers. Specifically, billing vulnerabilities and API key revocation delays expose users to immediate financial and security threats that platform policies do not yet fully address.