What happened
Incident overview / Railway and its customers
- 1
Customer secrets are exposed
October 20–21, 2026Railway found unauthorized reads of build histories and decryption of some customer credentials. It reset affected tokens and removed the intruder from its systems.
- 2
Customers incur unauthorized AI charges
After the intrusionCustomers supplied examples of unauthorized AI requests. Railway named Amazon Bedrock among providers with recorded financial impact and reported that AWS had begun issuing refunds.
Impact
Railway estimated about $740,000 in unauthorized AI charges across providers. The affected credentials included linked-account tokens for 1,852 users.
Could this have been prevented?
Limit which models an application can use
In a covered AWS member account, a service control policy can deny Bedrock inference on models the application does not need. A stolen credential could still run an allowed model and incur charges. Revoke an exposed credential even if model restrictions are in place.
Incident control assessments
- 0
- Prevented
- 1
- Contained
- 1
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Incident overview / Railway and its customers
1 · Customer secrets are exposed
1 Not covered2 · Customers incur unauthorized AI charges
1 ContainedKey facts
- Railway dates the intrusion to October 20–21, 2026.
- The intruder read the histories of 353 builds.
- Railway reset linked-account tokens for 1,852 users.
- Railway found no logins to its own customer accounts using the stolen credentials.
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Incident Report: October 21st Security AttackRailway · Nov 10, 2026 · Vendor reportThe timeline and customer impact sections establish the unauthorized reads and decryption. Attack intent and Unauthorized token spend identify customer-reported requests, the multi-provider estimate and Bedrock refunds.
- 2.Preliminary Report: Oct 21st Security IncidentRailway · Oct 22, 2026 · Vendor reportInitial timeline, containment and uncertainty before the final investigation.
- 3.API keysAmazon Web Services · undated · AdvisoryBedrock accepts short- and long-term API keys, with different lifetimes and permissions.
- 4.Prerequisites for running model inferenceAmazon Web Services · undated · AdvisoryIAM can limit model invocation permissions to specific supported models and resources.
- 5.Service control policies (SCPs)Amazon Web Services · undated · AdvisoryDescribes member-account coverage and exceptions for the management account and service-linked roles.
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.