Railway says AI keys stolen in its October breach ran up $740,000 in charges
App host Railway says an attacker who broke into its build system in October stole customers' AI provider keys and ran up an estimated $740,000 in charges, and it will cover what providers do not refund.
Warden, covering security, outages and the rules for tech
, 3 min read
$740Kestimated unauthorized AI charges on stolen keys
What to know
- An attacker inside Railway's systems on October 20 and 21 decrypted stored variables in up to 1.85% of workspaces, about 100,000, and took linked-account tokens for 1,852 users.
- Railway estimates about $740,000 in unauthorized charges on stolen AI provider keys; AWS and Google have started fraud refunds, while OpenRouter does not refund.
- Railway has filed a complaint with the FBI's Internet Crime Complaint Center (IC3) and moved every build into its own short-lived virtual machine.
Railway, a cloud platform developers use to host apps and databases, says an attacker who got into its build system in October stole customers' AI provider keys and used them to run up an estimated $740,000 in charges. The figure is in Railway's full incident report, published November 10, according to The Hacker News and iTnews. Neither outlet says how Railway arrived at the estimate.
What the attacker reached
Railway lets customers store secrets such as API keys as environment variables that their apps read at runtime. According to both reports, the attacker decrypted those variables in at most 1.85% of workspaces, which Railway puts at about 100,000 as an upper limit. The attacker also took linked-account tokens for 1,852 users, including GitHub OAuth tokens and keys customers had supplied for Railway's agent features. Railway says it reset all of them.
The queries the attacker ran were aimed at AI provider keys, The Hacker News reports, though some cloud, database and payment credentials matched them too. iTnews says the attacker favored customers with many stored variables.
Railway says it found no logins using the stolen credentials and no sign that customer databases or code repositories were accessed. The attacker did read 353 build histories, and iTnews reports it also got a copy of Railway's source repository.
How it happened
The intrusion started on October 20. The Hacker News gives Railway's timeline in UTC: the first exploit at 07:28 that day, the on-call engineer paged at 13:58, and a misconfigured cloud identity (OIDC) binding exploited at 03:26 on October 21. That binding led to a credential that could decrypt customer variables. Variables were pulled between 04:07 and 05:21, and the attacker was fully removed at 09:24 on October 21.
iTnews says a build credential was accepted by an internal workflow system, which is how the attacker moved from the build system into the rest of Railway. Railway described its response as a "contest of agent speed," according to iTnews.
Who pays
The stolen keys belonged to customers, so the charges landed on customers' accounts with each AI provider. Railway's update on October 26 said it could not refund that spending, The Hacker News reports. The later report gives a provider-by-provider picture, according to both outlets:
- AWS Bedrock and Google Gemini have started refunds under their fraud policies.
- OpenAI is reviewing claims case by case. Customers file with OpenAI and send the case ID to Railway support.
- Anthropic is still in talks, and Railway is covering costs in the meantime.
- OpenRouter does not refund, so Railway is covering those charges.
iTnews reports Railway said it would cover losses the providers do not refund.
What Railway changed
Every build now runs in its own short-lived virtual machine, and build credentials can push only one specific image, The Hacker News reports. Railway says every weakness the attacker used is closed and that it sent reset requests to more than 100,000 users as a precaution.
Railway has not named the attacker. It has filed a complaint with the FBI's Internet Crime Complaint Center (IC3), and a forensics firm is still investigating. Railway believes resale services used the keys to train other models on AI output, known as distillation, but has not established whether the attacker ran those services. It plans to publish indicators of compromise once law enforcement agrees.
Why it matters
A stolen AI key is now a direct line to someone else's bill. One break-in at a hosting platform turned customer secrets into hundreds of thousands of dollars in compute within days, and whether victims got their money back depended on each provider's fraud policy. For teams that keep AI keys on a hosting platform, the incident shows those keys carry the same risk as payment credentials.
For background, see our explainers on software supply chain attacks, how to read a security advisory and why companies publish post-mortems.
Sources
- Railway Says Stolen Customer AI Keys Led to $740,000 in Estimated Unauthorized Usage Charges, The Hacker News, November 11, 2026
- App hoster Railway attackers ran up US$740,000 on stolen AI keys, iTnews, November 11, 2026
Internet, outages and securityAI chips and data centers
Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.