Observed incident

Railway breach exposes stored secrets in an AI-key targeting campaign

Reported

Published here

What happened

Railway reported unauthorized internal access and decryption of some customer secrets. Preliminary forensics identified up to 1.85% of workspaces, approximately 100,000, with environment variables potentially compromised.

Parties and reporting

AI tools and systems: Railway build and CI systems, Stored environment variables, AI-provider and platform API keys

Outcome

Railway says attacker access ended October 22 and rotated credentials. Builds resumed; forensic investigation continued. An October 24 clarification says additional provider revocations were precautionary and inadvertently included replacement keys, rather than proving a new compromise.

Operating lesson

Rotate exposed secrets, review provider usage, separate build-system credentials and restrict secret decryption privileges.

Evidence and limits

AI assistance and distillation motive are Railway’s preliminary assessment; neither a specific attacking model nor successful distillation is established. Potentially affected workspaces are not confirmed theft totals. The preliminary report’s header predates its status-page publication announcement.

Sources