Railway details breach that generated $740,000 in fraudulent AI usage

Railway, a widely used platform for deploying apps and AI agents, published a detailed incident report on November 10, 2026, describing a late-October breach in which attackers exploited a build-isolation escape and a misconfigured OIDC binding to steal customer secrets. The attackers specifically hunted for AI provider API keys, which they used to generate approximately $740,000 in unauthorized AI usage.

What the report reveals

• The report was published on November 10, 2026, and was authored by four Railway engineers.
• The attack ran from October 20 at 07:28 UTC to full eviction on October 21 at 09:24 UTC, about 26 hours, with detection at October 20, 13:58 UTC.
• Attackers accessed 353 builds and ran direct database queries to decrypt and exfiltrate secrets, specifically targeting AI provider API keys.
• Up to approximately 100,000 workspaces, or 1.85% of Railway's roughly 5.4 million total workspaces, were potentially exposed.
• 1,852 users had linked OAuth or agent tokens compromised, including GitHub tokens.
• The stolen AI credentials generated approximately $740,000 in unauthorized AI usage.

What it means for people building with AI

The incident is a concrete supply-chain security warning for anyone running AI coding agents or apps on third-party platform-as-a-service infrastructure: developers' AI provider keys were specifically targeted and misused at real financial cost. Railway links the motive to credential resale and AI-model-distillation markets, framing the breach as part of a wider wave of attacks researchers call "Black October" against AI development platforms.

Remediation and next steps

• Railway's remediation includes dual-layer build isolation, a rebuilt CI/CD pipeline requiring 2FA or biometric approval, centralized encrypt/decrypt audit logging, a shift to short-lived just-in-time credentials, and a full rotation of credentials and PKI.
• Affected users are urged to rotate all credentials linked to their Railway workspaces.
• Railway has set up a reimbursement process for unauthorized AI charges and says it will cover costs that AI providers decline to refund.

Sources

• Incident Report: October Security Incident (Railway, November 10, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.