Railway says an attacker decrypted environment variables in up to 100,000 workspaces

Platform-as-a-service provider Railway published a preliminary incident report on October 22, a day after disclosing anomalous activity around the systems that store customer environment variables. Railway says the attack window opened on October 20 at 13:58 UTC with an attempted privilege escalation in build infrastructure. Between 04:24 and 05:27 UTC on October 21, the attacker queried encrypted environment variables, obtained CI credentials, and decrypted artifacts holding further API access. Logs show the attacker searching for named AI provider API keys. Railway estimates that at most 1.85% of workspaces, about 100,000, had variables potentially exposed, and it sent broader key lists to upstream vendors as a precaution. It revoked all Railway API tokens, rotated bucket credentials, and confirmed on October 22 that the attacker no longer had internal access. The entry point has not been disclosed.

Why it matters

Encryption at rest did not help once the attacker reached the CI identity that could decrypt. Railway believes it is one of a series of attacks on platform companies aimed at AI inference keys, the same target as the Zeabur and OpenCode incidents. Static API keys stored by a third party are only as safe as that party's least-protected credential.

Rotate every variable set on Railway before October 21, especially AI provider keys and private registry credentials, and recreate CLI and CI tokens. Remove public network access to databases. Review GitHub audit logs, installed apps and registry logs such as Docker Hub. Railway says it cannot refund AI spend made with stolen keys, so set spend caps with providers. Where providers support it, move to short-lived tokens through workload identity federation.

Source: Railway