What happened
Railway, a cloud platform for deploying apps, said attackers were inside parts of its systems from October 20 to 22, 2026, reached its CI environment and decrypted a subset of customers' environment variables; it says the goal was AI provider API keys for model distillation, that at most 1.85% of workspaces (about 100,000) were potentially affected, and that it asked GitHub, GitLab, Anthropic, OpenAI, OpenRouter, Stripe and others to revoke keys. Days earlier, TechCrunch reported on October 19 that OpenCode, an open-source AI coding agent, had its hosted OpenCode Console breached, exposing AI provider keys stored by 1.4% of Console users, with some reporting unauthorized charges above $1,000. Railway recommends that every customer who signed up before October 21 rotate the keys they store there. (Railway: Preliminary Report: Oct 21st Security Incident, Oct. 22, 2026)
Note on figures: The 1.85% and roughly 100,000 workspace figures, the attack timeline and the distillation motive are Railway's own preliminary findings; its investigation is ongoing and it has not ruled out access to source code. The 1.4% figure is OpenCode's, and the charge amounts are user reports cited by TechCrunch.
What it means for Canada: our assessment
Assessment: neither company names Canadian customers. Canadian startups, agencies and product teams that host apps on Railway or keep provider keys in OpenCode Console are exposed, because both platforms are used worldwide and the stolen secrets (OpenAI, Anthropic, Stripe, GitHub and database keys) open whatever those keys can reach, including customer data and billing.
- Sectors
- SaaS and AI startups · Fintechs · E-commerce and retail · Marketing and agencies · Universities and research labs
- Tools involved
- Railway · OpenCode Console · OpenAI API · Anthropic API · OpenRouter · Stripe API keys · GitHub and GitLab tokens · Docker Hub
- Canadian rules
- PIPEDA: if stolen keys were used to reach personal information (for example a production database or Stripe customer records), report breaches with a real risk of significant harm to the Privacy Commissioner, notify affected people and keep a breach record
Quebec Law 25: record every confidentiality incident and report serious ones to the Commission d'accès à l'information
OSFI B-13: federally regulated financial institutions must manage third-party and cloud technology risk, including developer platforms holding production secrets, and report material incidents to OSFI
Provincial health privacy laws (for example Ontario's PHIPA): custodians must report breaches if a compromised key gave access to health information
Assessment, not documented exposure. Neither Railway nor OpenCode publishes customer locations. The exposure is an assessment: both are self-serve global platforms popular with small software teams, and Canadian teams storing secrets there face the same rotation and abuse risk as any other customer.
Takeaways for Canadian businesses
- BusinessIf you used Railway before October 21, 2026, revoke and reissue every sensitive environment variable stored there (AI provider keys, Stripe keys, database passwords, cloud credentials), as Railway recommends, and remove public network access to Railway-hosted databases.
- BusinessIf you kept OpenAI, Anthropic, Google or other provider keys in an OpenCode Console workspace, revoke and replace them with the provider directly, then review usage and billing from October 16 onward.
- BusinessCheck your GitHub organization audit log and authorized OAuth apps, plus Docker Hub and other registry audit logs, for activity you did not start since October 20, as Railway advises.
- BusinessSet hard monthly spend limits and per-project keys in the OpenAI and Anthropic consoles so a stolen key cannot run up large charges, and report a confirmed compromise to the Canadian Centre for Cyber Security (cyber.gc.ca or 1-833-292-3722).
For the Canadian AI sector
The attackers were after AI keys specifically, which turns every place a Canadian AI startup parks an OpenAI or Anthropic key into a target and a billing risk. Expect more vendors to demand scoped, short-lived keys and spend limits, and Canadian AI firms to be asked by customers where their model keys live.
Written by AgentsNorth from the source below. General information, not legal or security advice.