FULL STORY

Railway Breach: The AI Keys Supply-Chain Attack

Railway disclosed unauthorized access to encrypted customer environment variables, with attackers targeting admin workspaces to steal AI API keys. The investigation and fallout continue to unfold.

2026-10-22 ~ 2026-10-24 · 2 episodes · 25 posts

Episode 1 · Railway Breach Exposes Encrypted Env Vars, Rippling Through AI Inference Supply Chain (2026-10-22, 23 posts)

Railway, a cloud deployment platform, disclosed a major security incident on Oct 21: it detected anomalous activity in the system storing customer environment variables, and an unauthorized third party may have accessed encrypted variables, some of which were decrypted. Railway paused all deployments, told all users to rotate every key on the platform, and said the attackers' real goal was to reach AI inference providers' API credentials through the platform.

Confirmed

  • Railway's disclosure states the breach was detected on the morning of Oct 21; environment variables may contain API keys, database and registry credentials. Deployments were halted and users told to rotate all credentials; per m2, Railway team member championswimmer confirmed builds are stopped and the team is working on recovery.
  • Per m1 relaying tayvano's timeline: builds paused at 07:24, API/CLI/Dashboard went down at 09:28, and users were only told to rotate keys roughly 12 hours after the outage began, drawing criticism of the slow response.
  • User varunram first warned that attackers were actively draining API keys and confirmed an ongoing attack.
  • Real losses have occurred: per m3, a Reddit user lost $80 after a Claude API key stored in Railway env vars was stolen; Anthropic caught it and disabled auto-recharge. Per m20, developer dobroslavdev reported an OpenRouter key stored only on Railway was drained.
  • Supply-chain fallout: per m14, OpenRouter auto-disabled affected users' keys, disrupting SaaS developers relying on it; per m13, open-source tool Postiz suspended all customers after Anthropic and Stripe notified it of leaked keys; Randy Olson publicly urged credential rotation.
  • Per m18/m19, the same day saw a LiteLLM supply-chain attack and an OpenCode hack, suggesting a dense wave of attacks on AI infrastructure.
  • Per m17, an alleged Railway employee's "last day" post admitting customer keys fell into attackers' hands appears to be a joke resignation post, not a credible source.

Unconfirmed

  • Railway's claim that this is part of a broader campaign targeting platform companies to infiltrate AI inference providers remains to be verified.
  • dejavucoder's joke that a rogue AI agent cluster did it is pure speculation; the full intrusion path has not been disclosed.

Why it matters

  • Decrypted environment variables mean credential-level defense in depth failed; downstream database, registry and service credentials could be cascadingly exposed. OpenRouter's proactive key disabling and Postiz's shutdown show the leak rippling through the AI inference supply chain to SaaS products and their customers.
  • The 12-hour gap before the rotation warning may have amplified losses and exposed weaknesses in hosted platforms' incident response.
  • If the same-day LiteLLM and OpenCode incidents are part of one chain, attackers are systematically targeting the AI infrastructure supply chain.

3 more related posts →

Episode 2 · Railary Discloses Breach Targeting Admin Workspaces and AI API Keys (2026-10-24, 2 posts)

Railway disclosed a sustained, targeted attack discovered on October 20, aimed at stealing AI API keys for distillation, and later notified workspace admins that some of their workspaces were linked to the threat actor.