Cloud infrastructure startup Railway disclosed that attackers used stolen customer API keys for AI services to generate roughly $740,000 in unauthorized usage charges before the activity was caught and cut off.
The company said compromised credentials, rather than any flaw in Railway's own platform, allowed the unauthorized usage to run up such a large bill before detection. Incidents of this kind have become increasingly common as more companies wire AI API keys directly into cloud deployment pipelines, where a single leaked credential can be used to rack up enormous charges in a short window before anyone notices.
The disclosure comes amid a broader run of credential-theft incidents affecting AI infrastructure providers this month, as attackers increasingly target the application programming interface keys that let software services call AI models, rather than going after the underlying AI systems themselves. Security researchers say stolen AI credentials are increasingly valuable precisely because usage-based billing can turn a single leaked key into a large and fast-moving financial loss for the victim.
Railway said it has taken steps to prevent a recurrence, though it has not detailed publicly how the keys were initially obtained. The incident underscores a growing financial dimension to AI-related security incidents: beyond data exposure, companies increasingly face direct monetary loss when API credentials for metered AI services fall into the wrong hands.