Railway favicon

Railway outage on 2026-10-24

railway.com
October 2026 24 Minor incident

On 24 October, Railway published a single notice that ran for 36h 1m, clarifying that a wave of AI provider key revocations was not a new attack. If OpenAI or Anthropic revoked a replacement key you'd already rotated after the 21 October incident, that was a side effect of Railway's own cleanup process, not a sign your new key was compromised.

Started

13:57 UTC

Duration

Lasted 1d 12h

Source

IsDown

Next time

Railway is degraded. Is your product working?

An incident on their side does not always mean an outage on yours, and the only way to know is to already be watching your own endpoints. Set that up in under a minute and you will be notified the next time your own site is affected. And there will be a next time.

50 monitors for your own websites. No card.

What happened?

Railway logged one incident on 24 October, starting at 13:57 UTC and staying open for 36h 1m. The single update posted was framed as a clarification rather than a fresh problem: Railway stated plainly that it was "not currently under attack" and that the API key revocations some of you were seeing were precautionary, tied back to the 21 October incident, not a new breach. According to Railway, the attacker's access was fully removed on 22 October at 03:32 UTC. The complication arose on 23 October, when AI providers asked Railway for another batch of potentially affected keys. Railway says it could not filter that batch by when keys were added, so newer keys, including ones you'd already created as replacements, got swept in and revoked by OpenAI and Anthropic. Railway was explicit that this did not mean your replacement keys were compromised, and told you to expect separate communications from the providers themselves about any action needed. The guidance given was straightforward: if a provider revoked your key, create another replacement, update the Railway environment variable, and redeploy the affected services. If your keys were still working, Railway said no further action was required. No components were listed as affected in this record, and no user reports were filed during the window. Only one update was posted for the full 36 hour span, and Railway's note closes with an apology for "the confusion and additional disruption."

Learning

This incident did not fail at the level your uptime checks would catch. Your Railway services may have kept responding normally throughout, while the actual breakage was happening one layer up, in the credentials your code depends on to talk to OpenAI or Anthropic. A key revocation doesn't take your endpoint down, it takes a downstream integration down, and that distinction matters because pinging your own URL tells you nothing about whether your AI provider calls are about to start failing with authentication errors. Railway's own history shows this kind of thing isn't rare for them: 31 incidents in the last 90 days, with a median duration of 1 hour 25 minutes, though this one ran far longer at 36h 1m. Our own 30 day figures for Railway show 16 incidents with a median of 1h 21m, which is the normal range, this one being an outlier worth remembering. The lesson here isn't about uptime at all, it's about knowing when a provider-side cleanup action can ripple into your own credentials without any outage on your side triggering it. Your monitoring covers your half, UptimeRobot now watches the provider's half too, so next time you know which side broke without guessing.

4.7
stars out of 5
284+ reviews on

Start monitoring in 30 seconds.

There's nothing to install. No credit card required. 50 monitors for free.