Railway's preliminary incident report describes a targeted attack on its build infrastructure beginning October 20 aimed at stealing AI provider API keys to perform model distillation attacks, with an unauthorized party potentially accessing decrypted environment variables containing API keys and other credentials.
Railway froze deployments, worked with Anthropic, OpenAI, OpenRouter, and others to revoke compromised keys, and told customers to rotate every secret. User accounts were not accessed, though source code exposure has not been ruled out.