Railway Build-System Breach Exposes Customer Secrets, Drives $740,000 in AI Charges
Sector Threat Intelligence · 17 sources · Updated 3d ago · First seen Nov 10, 2026

Railway disclosed an October 20–21 intrusion in which an attacker chained a partial build-isolation escape, excessive trust in a stolen build credential, and a cloud OIDC configuration error to reach production infrastructure and customer secrets. Access to stored build histories and internal source code helped the attacker identify additional attack paths, using AI agents even after the initial entry point was blocked. The attacker abused legitimate decryption permissions to recover some encrypted environment variables, with database queries specifically targeting AI-provider keys. Railway estimated approximately $740,000 in unauthorized AI-service spending. Variables in roughly 100,000 workspaces, or up to 1.85% of workspaces, could have been decrypted; that figure is an exposure upper bound, not a confirmed count of workspaces whose secrets were stolen.
Railway revoked credentials, rebuilt internal systems, reset linked-account tokens for 1,852 users, and restored builds across all plans by October 22. The company reported no evidence of customer database access or Railway account logins using stolen credentials and committed to covering breach-related charges where providers decline reimbursement. Customers should rotate potentially exposed secrets and review provider usage for unauthorized activity. The attack underscores the need to isolate build workloads, minimize build-credential privileges, validate cloud trust policies, and test runtime monitoring and outbound network restrictions.


