Zero Day Room
Live
Vulnerabilities

Railway breach exposed AI credentials in two-day exploit chain

Railway confirmed a Black October breach where attackers exploited two vulnerabilities, stole AI credentials, and generated $740,000 in fraudulent usage, potentially exposing 100,000 workspaces.

Railway confirmed a Black October breach where attackers exploited two vulnerabilities, stole AI credentials, and...

Railway confirmed on October 22 that hackers stole customer authentication secrets and generated approximately $740,000 in fraudulent AI usage, potentially exposing credentials for up to 100,000 workspaces. The attack began on October 20 at 07:28 UTC when threat actors exploited a previously unknown flaw in the company's multi-tenant build infrastructure to escape filesystem isolation. Attackers obtained a build server credential used to push images and access internal systems, triggering an alert six and a half hours later at 13:58 UTC.

Scope of compromise

The breach exposed decrypted customer environment variables for at most 1.85% of workspaces, or about 100,000, described as the maximum potentially affected count. Attackers read 353 build histories, including the secrets and variables passed to those builds, and copied Railway's internal monorepo. The company found no evidence hackers downloaded customer Git repositories or accessed customer databases. Researchers link the incident to a broader campaign called Black October that also targeted AI developer platforms Northflank, OpenCode, Warp, and Lightning AI.

Attackers specifically searched for AI provider credentials, likely for resale or to power AI model distillation attacks. They also looked for cloud, database, and payment service credentials. Railway believes the attacker wanted AI credentials to sell for model distillation, with the stolen keys used by services that relay or resell AI access.

Technical attack vectors

On October 21, the attackers used a misconfigured OIDC binding to obtain short-lived cloud credentials. They then used those credentials to decrypt customer environment variables, including AI provider keys, via a Key Management Service and an orchestration machine. This decryption activity occurred in two distinct windows: between 04:42 and 05:14 UTC and again between 06:36 and 07:15 UTC.

The attacker accessed cloud, database, and payment service credentials via a database query. Railway stated that no customer logins using stolen credentials were found. The company described a 26-hour contest between its engineers and threat actors who used AI-powered agents to probe and exploit weaknesses in parallel.

Response and remediation

Railway fully removed the attacker from its systems at 09:24 UTC on October 21. The company patched the original build system flaw and moved all users to a new build system within an hour. Every build now runs on its own short-lived virtual machine with a credential that can push only one specific image. The company also hardened system routes in parallel as attackers tested many options at once.

All encrypt and decrypt operations are now logged. Railway reset 1,852 linked-account tokens, including GitHub OAuth tokens and AI agent keys, and notified affected users. More than 100,000 users received reset requests due to key revocations from providers. The company advised customers to remove public network access to databases and to check GitHub and registry audit logs. It emphasized that keys must be revoked with the provider that issued them, as submitting a charge report does not automatically revoke a key.

Financial impact and refunds

The unauthorized spend on AI keys stolen from Railway customers reached approximately $740,000. Refund policies for the fraudulent charges vary significantly by AI provider.

Railway stated customers are not responsible for usage of compromised keys but must report unauthorized charges via a dedicated form. The company is working with affected providers to reimburse customers and pledged to cover charges that providers decline to refund.

Related coverage

More from Vulnerabilities