ACN · Presented by Levr
ACN WIRE

Security

Stolen AI keys run up $740,000 at Railway

One of at least nine AI platform breaches in October

Written with AI assistance, reviewed by the ACN desk.

Why it matters

If a platform holding your AI keys is breached, the charges land on your account.

The quick hit

  • Railway says attackers stole customers' AI provider keys and ran up an estimated $740,000 in charges.
  • Up to about 100,000 workspaces could have had secrets decrypted, Railway says.
  • Wiz counts at least nine AI developer platforms that disclosed incidents in October.

Railway, a cloud platform that builds and hosts apps for developers, says attackers stole AI provider keys its customers stored on the platform. Unauthorized spend on those keys was estimated at about $740,000, according to Railway’s incident report, published on Tuesday, November 10.

The intrusion began on Tuesday, October 20, when the attacker exploited a previously unknown flaw in Railway’s build infrastructure and partly escaped its isolation. From there the attacker moved across internal systems, reached a machine that could decrypt customer variables, and searched Railway’s database for AI keys by name. Railway says the attacker was fully removed at 09:24 UTC on Wednesday, October 21.

Railway says at most 1.85% of workspaces, about 100,000, could have had environment variables decrypted, and calls that an upper bound. Linked-account tokens for 1,852 users, including bring-your-own agent keys and GitHub OAuth tokens, were reset. Railway found no logins using stolen credentials and no sign that customer databases were accessed.

Railway believes AI keys were the main target and that they were passed on for distillation: training other models on a model’s answers. It also says the attacker used AI agents, describing the response as “a contest of agent speed.”

Railway was not the only target. In its “Black October” research, security firm Wiz counts at least nine AI developer platforms that disclosed incidents in October, including Fireworks AI, OpenCode and Warp, which Wiz says reported signs of an “agent swarm” attack. Wiz says the attacks hunted for stored AI credentials, a pattern it calls LLMJacking, and that one platform, Zeabur, is shutting down after its breach. Wiz has not tied every incident to the same attacker.

Who pays depends on the provider. Railway says AWS and Google are refunding under their fraud policies, and OpenAI reviews claims case by case. Railway is covering Anthropic charges while talks continue, and will cover all OpenRouter charges because OpenRouter enforces its no-refund policy.

The Hacker News notes that the report does not say how the $740,000 figure was estimated. Railway advises customers to rotate any secret stored on the platform before October 21, with the provider that issued it, and to report charges they did not make. Filing a report does not revoke a key.

Sponsor

Sources

  1. Railway (incident report) · Nov 10, 2026
  2. Wiz Research · Oct 30, 2026
  3. The Hacker News · Nov 11, 2026

Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.

From our sponsors

Sponsor

Sponsor

Sponsor

Sponsor