The Railway breach let attackers use stolen customer credentials to run up roughly $740,000 in unauthorized AI charges, the cloud hosting provider said on November 10. Railway is working with AI providers to reimburse affected customers and says it will cover any charges the providers decline to refund.

Up to 100,000 workspaces faced potential exposure

The attack began on October 20, when intruders gained access to Railway’s build infrastructure and obtained an internal credential. They later accessed stored build histories, copied Railway’s internal source code and obtained credentials that could decrypt customer environment variables. Developers use these variables to store settings and secrets, including the API keys their applications need to access paid services.

Railway estimates that the attackers could have reached decrypted variables for at most 1.85% of its workspaces, or about 100,000. That figure sets an upper limit on the exposure. The company has yet to establish how many workspaces actually had secrets stolen.

The Railway breach also exposed linked-account tokens belonging to 1,852 users. These included GitHub OAuth tokens and AI agent keys. Railway reset the tokens and notified those users. Its investigation found no evidence that the attackers downloaded customers’ Git repositories or accessed their databases.

Attackers sought keys for paid AI services

Over roughly 26 hours, the attackers used AI agents to investigate several potential weaknesses at once, according to Railway. Meanwhile, its engineers worked to close those routes. The company said it removed the attackers from its systems on October 21 and has since fixed the vulnerabilities they exploited. On October 22, it publicly urged customers to rotate their secrets.

The attackers’ searches focused on AI provider credentials, although some also targeted secrets for cloud platforms, databases and payment services. The billing records showed a direct consequence for customers: the stolen keys let someone else consume AI services at their expense.

Railway suspects the attackers wanted keys to resell or to use for model distillation, a process that uses one model’s outputs to train another. The company linked the incident to a broader series of attacks on AI development platforms, including Northflank, OpenCode, Warp and Lightning AI. The attackers’ ultimate purpose for the stolen keys remains unconfirmed.


0 responses to “Railway breach leads to $740,000 in fraudulent AI charges”