# Northflank is the ninth AI dev platform breached in October > Attackers are taking AI keys from the platforms that host them. In two cases Wiz traced, stolen keys sat quiet for days, then ran up an estimated $100k-plus. - Published: 2026-10-31 - Author: 28creation Desk - Format: News - URL: https://28creation.com/articles/northflank-ninth-ai-platform-breach-october/ ## The short read **Two AWS keys stolen in a platform breach sat nearly idle for days, then ran up over $100,000, Wiz estimates.** - Northflank told all customers on October 30 to assume their environment variables were decrypted. Wiz counts nine AI developer platforms that disclosed incidents this month. ### What you can do with this > Revoke every AI key stored on a hosting platform with the provider that issued it, then check usage back to mid-September. A trickle of calls you did not make is the early warning. --- Northflank, a platform that runs developers' apps, told every customer on October 30 to "assume all environment variables have been decrypted" and rotate them, [per its status page](https://status.northflank.com/cmvv0rn5a092f14rutwccodpp). Users reported their AI keys being used. Hours later, [Wiz counted Northflank](https://www.wiz.io/blog/black-october-for-ai-developer-platforms) as the ninth AI developer platform to disclose an incident this month. If your team stores an AI provider key on a hosting or AI platform, this campaign is after it. The useful finding is timing. In the two cases Wiz traced, the keys were tested within 30 minutes, then sat nearly idle for four days or more before heavy abuse began. ## Nine platforms, one target Wiz lists the disclosures in order: Fireworks AI (October 6), Andromeda (October 9), Zeabur (October 14), Lightning AI (October 15), OpenCode (October 16 to 17), Railway (October 20), Warp (October 23), CommandCode (October 28) and Northflank (October 30). It found "no single vulnerability or dependency linking these attacks." The entry points differ: credential access at Fireworks, Zeabur and Lightning, build infrastructure at Railway, a vendor dependency at Warp. The goal does not differ. Every public report, Wiz writes, shows attackers "explicitly hunting stored environment variables," mainly those that grant AI access. A [Cloud Security Alliance note](https://labs.cloudsecurityalliance.org/research/csa-research-note-inference-provider-platform-chain-campaign/) from October 23 had already grouped the first incidents as one campaign aimed at environment variables and AI keys. Why AI keys? They turn into money fast. Wiz found two resale operations testing the stolen keys; both sell discounted access to frontier models, and one takes payment in USDT, a dollar-pegged cryptocurrency. The damage is real for the platforms too. [Railway](https://blog.railway.com/p/preliminary-report-october-security-incident) says at most 1.85% of its workspaces had variables potentially exposed, and that it cannot refund spend made with compromised keys. [Zeabur](https://zeabur.com/blogs/zeabur-is-shutting-down) is closing, with December 31 as the final shutdown date; it named two security incidents in the second half of 2026 among its reasons. ## The keys sat quiet before the flood Wiz followed two enterprise AWS keys stolen in one of the vendor breaches. Both got a single test call within 30 minutes of the breach. Then came 14 to 120 calls a day from one small client, a job checking that the keys still worked. In Case A, the flood started four days later: about 800,000 calls to Amazon Bedrock, AWS's model service, over 30 hours, about 90% of them successful. Case B's key turned up in a reseller pool seven days after its test. The next day, one address made about 1.7 million calls in 13 hours, over a million of them successful. At list price, Wiz estimates, the two keys alone ran up more than $100,000. Case B's key was locked to US-only routing, so most failed calls hit regional quotas. The limit throttled many calls and still let over a million through. Case A's key allowed global routing, and only about 6% of its calls were throttled. ## What to do now 1. **Rotate at the source.** Revoke each key with the provider that issued it, not only on the platform. Northflank and the CSA both say this; removing a secret from a dashboard does not revoke it. 2. **Check usage back to mid-September,** as the CSA advises: unfamiliar models, regions or volume. A trickle of daily calls you did not make is the warning Wiz saw days before each flood. 3. **On Railway,** rotate everything if you signed up before October 21, as the company recommends. 4. **Give each platform its own key,** with a spending cap and rate limit near normal use, so one breach costs one key. The trade-off is upkeep: more keys mean more to rotate. The quiet days in Wiz's cases are the case for doing it anyway. A key you replace this weekend is worth nothing at the resale sites. ## Sources - Northflank status page, security incident, October 30 to 31, 2026: https://status.northflank.com/cmvv0rn5a092f14rutwccodpp - Wiz, "Black October" for AI Developer Platforms, October 30, 2026: https://www.wiz.io/blog/black-october-for-ai-developer-platforms - Cloud Security Alliance, research note on the inference provider platform campaign, October 23, 2026: https://labs.cloudsecurityalliance.org/research/csa-research-note-inference-provider-platform-chain-campaign/ - Railway, preliminary report on the October security incident: https://blog.railway.com/p/preliminary-report-october-security-incident - Zeabur, Zeabur is shutting down, October 17, 2026: https://zeabur.com/blogs/zeabur-is-shutting-down - Runtime Wire, Northflank tells customers to assume all environment variables are compromised, October 30, 2026: https://runtimewire.com/article/northflank-tells-customers-to-assume-all-environment-variables-are-compromised-a