The Hacker News
Railway says stolen AI keys led to $740,000 in charges
Railway, a platform‑as‑a‑service provider, disclosed that stolen customer AI provider keys stored as environment variables on its platform were used to rack up an estimated $740,000 in unauthorised usage charges. It said the attacker decrypted variables in up to 1.85% of workspaces, recognising AI provider keys during database queries that searched for such credentials.
In an October update, Railway confirmed it could not refund customers for spend incurred with compromised API keys and directed affected customers to pursue refunds directly with the AI providers, while offering to escalate where appropriate.
Railway’s incident report also notes that the attacker decrypted additional data, including linked‑account tokens for 1,852 users (BYOK agent keys and GitHub OAuth tokens). All affected tokens were reset, with notifications sent to users, and no customer account logins were observed using the stolen credentials. Beyond AI keys, the intruder’s queries matched cloud, database and payment service credentials; however forensics showed no access to customer databases and no downloads of customer code repositories.
The attacker also read the stored histories of 353 builds, including the secrets and variables passed to those builds. Railway states the intruder was fully removed by 09:24 UTC on 21 October, after exploiting a previously unknown flaw to partly escape filesystem isolation and access a credential used by build servers; the breach prompted the company to segregate builds into short‑lived VMs and to patch the underlying weaknesses.
The company is cooperating with law enforcement (FBI IC3) and a forensics firm, and plans to publish indicators of compromise in due course. It notes that stolen credentials were used by services that relay or resell access to AI providers for use in distillation, but has not named a specific attacker.