Railway says stolen customer secrets led to about $740,000 in fraudulent AI usage
The cloud host says attackers combined an unknown flaw in its build system with a misconfigured trust relationship, and that up to 100,000 workspaces may be affected.
Evidence: Independent reports. Security stories run only with a named disclosure or independent reporting behind them.
Cloud hosting provider Railway has published an incident report on the breach it first disclosed on October 22. Hackers stole customer authentication secrets and used them to run up about $740,000 in fraudulent AI usage. Railway says the attack began on October 20. The attackers exploited a previously unknown vulnerability in its build infrastructure to break out of filesystem isolation and obtain an internal credential. They used that credential to reach stored build histories and copy Railway's internal source code, then studied the code for further weaknesses. A second vulnerability, an improperly configured trust relationship, gave them deeper access and the credentials used to decrypt customer environment variables. Railway says decrypted environment variables were exposed for at most 1.85% of its workspaces, about 100,000. It stresses that this is the maximum number possibly affected, not the confirmed number of workspaces whose secrets were stolen. The Hacker News reports that the stolen secrets included customers' AI keys. Railway had already told customers to rotate their secrets. The case shows that a breach at a hosting provider can quickly become a direct cost for customers whose stored AI API keys are abused.