Railway Hack Exposed 100,000 Credentials and Triggered $740,000 in AI Fraud

3 articles · Updated · Computerworld · Nov 12

Railway said attackers spent nearly 26 hours inside its systems after an Oct. 20 breach, decrypting a subset of customer environment variables tied to roughly 100,000 credentials, or at most 1.85% of workspaces.
Two flaws drove the intrusion: a previously unknown build-infrastructure vulnerability and an improperly configured trust relationship that let hackers access build histories, study Railway’s source code and obtain credentials used to decrypt customer secrets.
The stolen keys generated about $740,000 in fraudulent AI usage charges, with customers reporting unauthorized coding queries and model-training tasks; Railway said it will reimburse losses that providers do not refund.
Railway also reset linked GitHub and OAuth tokens for more than 1,850 users, worked with AI and cloud partners to revoke exposed keys, and urged customers to rotate all sensitive variables stored before Oct. 21.
The incident highlights how stolen AI credentials are becoming a resale commodity for model distillation and other high-volume workloads, pushing platforms toward short-lived credentials, tighter key scopes and spending caps.