On 21 October 2026, the cloud deployment platform Railway disclosed that an unauthorised third party had accessed systems that store customer environment variables and decrypted some of them. Railway's preliminary report of 22 October says the attack ran from 20 to 22 October 2026: a suspected privilege escalation in build infrastructure, followed by theft of CI environment credentials that were then used to decrypt a subset of encrypted variables. Railway says at most 1.85% of workspaces had some variables potentially compromised. It believes the attacker was after AI provider API keys for distillation attacks, and links the incident to a series of attacks on platform companies. Railway paused all deployments, revoked every login session, access token and GitHub App installation, and asked providers including OpenAI, Anthropic, GitHub and Stripe to revoke keys. Railway has not disclosed how the attacker first got in, and this page does not guess.
Key takeaways
- Railway disclosed on 21 October 2026 that an attacker accessed encrypted customer environment variables and, for some, "obtained the decrypted contents".
- According to Railway's preliminary report, the attacker used credentials taken from Railway's CI environment to decrypt a subset of variables. At most 1.85% of workspaces were potentially affected.
- Railway says early evidence shows the attacker wanted AI provider API keys "to perform distillation attacks", and it asked upstream vendors to revoke keys, touching more than 100,000 users.
- The initial access vector has not been disclosed. Railway says it is "not yet able to share the specific details and scale of the exploits" while its investigation continues.
- The identity lesson: encryption at rest protects secrets only as well as the machine credentials that can decrypt them. Once CI credentials with decrypt rights were taken, the encryption no longer stood in the way.
At a glance
| Organisation | Railway, a cloud deployment platform, and a subset of its customers (at most 1.85% of workspaces, according to Railway) |
|---|---|
| When | Attack window 20 October 2026 13:58 UTC to 22 October 2026 03:32 UTC; security disclosure 21 October 2026; preliminary report 22 October 2026 |
| Attacker | Unknown. Railway describes a "hostile external party" seeking AI provider API keys and links it to a series of attacks on platform companies; no group has been named |
| Entry point | Not disclosed. Railway's first finding was a suspected privilege escalation attempt in its build infrastructure |
| Identities abused | Railway CI environment credentials used to decrypt artifacts and environment variables; customers' stored API keys, AI provider keys, database and registry credentials |
| Impact | A subset of encrypted environment variables decrypted; at most 1.85% of workspaces potentially affected; deployments paused for about a day and a half; all sessions, tokens and GitHub connections reset |
| Category | NHI. Incident class: confirmed NHI breach (CI credentials used to decrypt customer secrets; root cause undisclosed) |
What happened
Railway hosts applications for developers and stores their configuration, including secrets, as environment variables. On 21 October 2026, chief executive Jacob Cooper published a security disclosure: "Earlier this morning, Railway detected anomalous activity involving systems that store customer environment variables." Railway said an unauthorised third party may have accessed encrypted environment variables and, for some, "obtained the decrypted contents." It said the incident looked like "part of a series of attacks recently targeting platform companies" aimed at "compromising AI inference providers", and that it had asked upstream providers to revoke keys proactively. All deployments were paused, and every customer was told to roll their environment variables and private registry credentials.
A preliminary report from Cooper and Angelo Saraceno on 22 October gave more detail. Monitoring paged on-call engineers at 13:58 UTC on 20 October about elevated latency, and the team found a suspected privilege escalation attempt in build infrastructure. A fix was deployed that afternoon, but by 19:53 UTC evidence showed the attacker could escalate access more widely and in parallel. More access routes turned up overnight. Early on 21 October Railway found the attacker had pulled secrets in encrypted form but could not yet decrypt them. Between 05:27 and 07:25 UTC, the attacker obtained CI environment credentials and decrypted artifacts holding further API access. Database evidence then showed unauthorised queries of encrypted environment variables, with a subset decrypted using the compromised credentials.
Railway notified vendors at 09:25 UTC on 21 October (its FAQ gives 09:35). It rolled all keys, then rolled them again after confirming the intruder had gone, rolled credentials for all storage buckets and GitHub App installations, and revoked all Railway access tokens and login sessions. The report says that "we can validate that at most 1.85% of all workspaces had some environment variables potentially compromised." Because keys were envelope encrypted, Railway says the query evidence cannot show exactly which keys in a given project were affected, so it is also contacting a larger group of high-risk customers. It says: "The attacker did not access Railway user accounts as part of this incident", though access to customer source code has not yet been ruled out.
On motive, Railway writes: "Early investigations tell us this attack was attempting to obtain AI provider API keys to perform distillation attacks." It does not say how the attacker first reached its build systems, stating that "we are not yet able to share the specific details and scale of the exploits." RuntimeWire, reporting the same day, also gave no root cause. Enterprise and Pro builds came back early on 22 October and normal deployments resumed for all tiers later that day. Follow-up guidance on 23 October explained that "API tokens and CLI sessions were revoked, so manual deploys from your machine or CI fail until you create new credentials."
Timeline
| Date | Event |
|---|---|
| 20 October 2026 | Monitoring flags abnormal latency at 13:58 UTC; a suspected privilege escalation in build infrastructure is found and an incident is posted on the status page. |
| 21 October 2026 | Railway finds encrypted secrets were pulled; the attacker obtains CI credentials and decrypts a subset of environment variables; vendors are notified; the security disclosure is published. |
| 22 October 2026 | Internal systems rebuilt with new credentials; builds restored; preliminary report published. |
| 23 October 2026 | Railway publishes recovery guidance for failed logins and deploys after mass credential revocation. |
How it happened: the identity attack path
- Initial access, not disclosed. Railway has not said how the attacker got in. The first thing it found was a suspected privilege escalation attempt in build infrastructure.
- Privileges widened inside the platform. Railway says the attacker could escalate access at greater scale across several parts of the system in parallel, and found further access routes overnight.
- Encrypted secrets taken. The attacker pulled customer secrets in encrypted form. At that point, according to Railway, it could not decrypt them.
- CI credentials stolen. The attacker then obtained credentials from Railway's CI environment and used them to decrypt artifacts that held additional API access.
- Variables decrypted and keys targeted. Using those credentials, the attacker queried and decrypted a subset of customer environment variables. Payload logs showed the specific API key names being sought, which Railway ties to AI provider keys.
Impact
- Confirmed by Railway: unauthorised queries of encrypted environment variables, with a subset decrypted; at most 1.85% of workspaces had some variables potentially compromised; no access to Railway user accounts.
- Not yet ruled out: access to customer source code, which Railway says its final report will address. Railway has not published how many secrets were decrypted or whether any have been misused.
- Precautionary: upstream vendors including GitHub, GitLab, Anthropic, OpenAI, OpenRouter and Stripe were asked to revoke keys, affecting more than 100,000 users according to Railway.
- Operational: deployments were paused from 20 October until 22 October; all login sessions, CLI sessions, API tokens and GitHub connections were revoked, and commits pushed during the pause were not deployed.
- Wider: Railway links the incident to a campaign against platform companies aimed at AI provider keys, the same kind of target seen at Zeabur and OpenCode in the same month. Railway has not said whether the incidents share an actor.
What this means for NHI and AI agent security
Railway encrypted customer secrets at rest, and that did slow the attacker: for a time, it held the ciphertext but could not read it. The decisive step was the theft of a machine identity, the CI environment credentials that had the right to decrypt. Encryption moves the risk from the data to whoever holds the keys, and in many platforms that is a build or deployment identity that is broad, long-lived and rarely watched. Pipeline identities deserve the same protection as production administrators, as our CI/CD Pipeline Identity Security Guide explains.
The target matters too. Railway says the attacker wanted AI provider keys for distillation, using stolen access to query models at scale. Platforms that host AI applications now hold large numbers of high-value AI keys, which makes them a single point from which many keys can be stolen at once. Customers cannot fix a platform's internal controls, but they can limit what a stolen key is worth, with spending caps, per-platform keys and fast revocation, as set out in our LLMjacking Guide.
Railway's response also shows what revocation at scale looks like: every session, token and GitHub installation reset, and vendors asked to revoke keys for more than 100,000 users. It was disruptive, but it is the right instinct when the decrypted set cannot be pinned down. The Leaked Credential Response Playbook covers how to prepare for it.
Recommendations
- Rotate every secret you stored on Railway. Railway recommends this for all users, especially anyone who signed up before 21 October 2026, even if your provider has already revoked a key. See our Secrets Rotation Runbook Template.
- Recreate revoked platform credentials with the same scope. Issue new project and workspace tokens, reconnect GitHub and update CI secrets, without widening the scope of the old tokens.
- Check AI provider and cloud usage for misuse. Review billing and usage logs for AI, cloud and payment keys stored on the platform, from 20 October onwards. See our LLMjacking Guide.
- Treat CI and build identities as privileged. Give build systems short-lived credentials, keep decrypt rights out of general CI environments and alert on unusual decrypt operations. See our CI/CD Pipeline Identity Security Guide.
- Cap the value of keys held by third parties. Use a separate AI provider key for each hosting platform, with spending limits and narrow permissions, so one platform breach does not drain a shared key.
- Prepare for mass revocation. Know where every platform-held secret is used and who owns it, so you can rotate within hours when a provider discloses an incident. See our Leaked Credential Response Playbook.
Frequently asked questions
What happened in the Railway security incident in October 2026?
Between 20 and 22 October 2026 an unauthorised party gained access to Railway's internal systems, stole credentials from its CI environment and used them to decrypt a subset of customers' encrypted environment variables. Railway says at most 1.85% of workspaces were potentially affected. It paused deployments, revoked all sessions and tokens, and asked AI and other providers to revoke keys.
How did the attacker get into Railway?
Railway has not disclosed it. Its preliminary report describes a suspected privilege escalation in build infrastructure and the later theft of CI credentials, but it says it cannot yet share the specific details of the exploits while forensic work continues. Any claim about the initial access vector is speculation until Railway's final report.
Do Railway customers need to rotate their secrets?
Yes. Railway recommends that all users roll the environment variables, API keys and private registry credentials stored on the platform, not only those it contacts directly. Because of envelope encryption, Railway says it cannot tell exactly which keys in a project were exposed, so rotating everything is the safe course.
Related NHI Mgmt Group resources
Zeabur second breach 2026 · OpenCode Console breach 2026 · PocketOS database deletion 2026 · CI/CD Pipeline Identity Security Guide · LLMjacking Guide
How NHI Mgmt Group can help
When a platform you rely on is breached, the work lands on your team: finding every key you gave it, rotating fast and checking for misuse. We help organisations map secrets held by third-party platforms, limit what each key can do and build the playbooks that make mass rotation routine. See our NHI and AI agent security training.
References
- Railway: Security Disclosure, Wednesday October 21st (21 October 2026)
- RuntimeWire: Railway pauses deployments after possible exposure of customer environment variables (21 October 2026)
- Railway: Preliminary Report: Oct 21st Security Incident (22 October 2026)
- Railway: Security Disclosure Recommendation: Recover Failing Deploys (23 October 2026)