Skip to newsletter content
Will Lassalle
The Governance Gap

Issue #34

Your AI API Key Is Now a Corporate Credit Card

AI credentials carry spending authority. Why Security, Technology and Finance need tested controls for unauthorized consumption.

The Governance Gap | Issue #34

THE TAKE

One of the more revealing cybersecurity stories in November has nothing to do with ransomware or stolen customer records. It involves attackers stealing access to something increasingly valuable: AI computing power.

On November 10, Railway released details of its October cyberattack, revealing that the attackers specifically targeted AI provider credentials stored in customer environments. The company estimated approximately $740,000 in unauthorized AI token usage. That is an estimate of consumption, not customers' final losses: Railway says it is coordinating refunds with providers and covering token-related losses where providers will not. Separately, Wiz Research documented stolen enterprise credentials being used for millions of attempted AI model calls. Those examples should not be assumed to be Railway customers. There's now an active market for stolen AI access, and criminals have figured out how to monetize it.

What concerns me is how easily this exposure can fall between traditional security and financial controls. Most organizations protect corporate purchasing cards, bank accounts, and privileged credentials through established processes. Yet developers routinely create API keys, connect them to applications, and authorize consumption-based services with far less oversight. My view is that boards need to start treating AI credentials as security assets that also carry spending authority. They provide access, but they can also authorize spending.


THREE THINGS BOARDS SHOULD KNOW

  • AI credentials have direct financial value: A compromised API key can allow attackers to consume expensive computing resources without stealing money from a conventional financial account.
  • Third-party platforms can expose downstream AI services: Credentials stored in development platforms, automation tools, and cloud environments may become accessible when those providers are compromised.
  • Traditional cloud monitoring may miss the financial impact: An organization can experience substantial unauthorized AI consumption before anyone recognizes the usage as malicious.

THE QUESTION I'M HEARING

"Aren't our existing cloud security controls already protecting these credentials?"

Not necessarily. Traditional cloud security provides an important foundation, and metered cloud spending risk predates AI. AI expands the services and credentials through which that exposure can grow. A valid credential can generate significant expense through perfectly legitimate API calls. From the provider's perspective, those requests may look authorized unless additional restrictions or monitoring are in place.

The starting point is understanding where AI credentials exist, who owns them, what services they can access, and what financial exposure they create. Organizations should apply least privilege, use managed secrets, restrict credential scope, and implement consumption monitoring with meaningful alerts and enforced spending caps wherever supported. An alert tells someone that usage has crossed a threshold; a cap can stop further consumption. Boards should know which controls actually stop spending, their limitations, and who responds when no hard cap is available. Development teams should not be able to create unlimited financial exposure simply by connecting an application to an AI service.

My recommendation is to bring Security, Technology, and Finance together on this issue. Include AI credentials in privileged-access reviews, monitor unusual consumption patterns, establish incident procedures for rapid revocation, and understand the provider's policies for disputed usage. Most importantly, test whether the organization can identify and contain unauthorized consumption before the monthly invoice arrives, and verify how quickly revocation and spending limits take effect.


THE BOTTOM LINE

An AI API key is no longer just another technical credential. It can represent direct access to expensive computing resources.

If your organization controls spending authority more carefully than AI credentials, you have a governance gap.

SOURCES