Railway, the cloud platform thousands of developers use to deploy apps and run AI workloads, has published its final incident report on the October breach — and the numbers are worse than the first disclosures suggested. Attackers who broke into its build systems decrypted environment variables in up to 100,000 customer workspaces, exposed linked-account tokens belonging to 1,852 users, and ran up roughly $740,000 in fraudulent AI charges on stolen keys.

The report, published on November 10, revises the timeline and scope of an attack Railway first disclosed in late October. It also reverses the company's earlier position on who pays for the fraudulent charges — Railway now says it will absorb the cost itself wherever an AI provider declines to issue a refund.

How the attackers got in

The attack began on October 20 at 07:28 UTC, when an intruder used a previously unknown flaw in Railway's build infrastructure to partly escape the filesystem isolation around a build and read files on the host virtual machine. The attacker seized a build credential that an internal Temporal cluster accepted as a trusted client, opening the door to the stored histories of 353 builds — secrets included — and to a copy of Railway's own internal source code.

Engineers revoked all static and active cloud credentials at 07:14 UTC on October 21 and halted customer builds twice during containment. The attacker was fully evicted at 09:24 UTC that day, about 26 hours after the intrusion began. The revised timings replace Railway's preliminary report, which had dated the attack from later on October 20.

What was taken — and what was not

Railway says decrypted environment variables could have been taken from a maximum of 1.85 percent of its workspaces — roughly 100,000, as an upper bound. The company stressed that the figure is a ceiling, not a confirmed count: its logs record which environments were exposed, not which individual variables inside them were decrypted. Linked-account tokens for 1,852 users, including GitHub OAuth tokens and bring-your-own-key agent keys, were exposed; Railway says all of them have been reset and the affected users notified.

Forensic logs show the attacker specifically searched for AI provider credentials, and Railway believes the keys were targeted for sale to be used in model distillation. The company's investigation found no sign that customer databases were accessed, that code repositories were downloaded, or that anyone logged into a Railway account using stolen credentials.

Railway will now pay the $740,000 bill

The most striking change in the final report is the refund reversal. In an October 26 update, Railway told customers asking about reimbursement that it was unable to refund them for spend performed with compromised API keys. The final report takes the opposite stance: Railway is working with AI providers on refunds and will cover the estimated $740,000 in unauthorized usage itself where providers do not reimburse customers.

The incident is part of a wider pattern in which AI credentials have become a prime target. The Fireworks AI breach and the OpenCode API-key incident involved the same kind of hunt for AI provider keys. For teams running AI workloads in the cloud, the lesson is a practical one: stolen keys do not just leak data, they can generate a large bill in hours, so spending limits, usage alerts and short-lived credentials are worth setting before an incident, not after.