The daily publication · November 2, 2026 · Technology
Your AI key is a wallet. Nine platforms that hold them got breached in one month

Listen
Narrated by Tabron · 7.4 min
Nine companies that build and host AI tools told their customers in October that they had been breached. The target was not customer data in the usual sense. It was the keys.
The idea
Cloud security firm Wiz published its count on October 30 and called it Black October. Researchers Benjamin Read and Rami McCarthy wrote the report. At least nine AI developer platforms disclosed security incidents between October 6 and October 30: Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode, and Northflank.
The common thread was what the attackers went looking for. These platforms store environment variables for their customers, and those variables hold secrets. The prize secret is an AI API key, the credential that pays for model calls. Wiz calls the abuse that follows LLMjacking: someone uses your key to run models, and you get the bill, or they resell the access to someone else.
The speed is the part to sit with. In two cases Wiz investigated, a stolen enterprise key was tested within 30 minutes of the breach at the vendor. The two cases together ran about 2.5 million Amazon Bedrock API calls, which Wiz estimates at roughly $100,000 in usage at list price.
The latest name on the list shows how it ends for a customer. Northflank disclosed its incident on October 30 after a threat alert from Wiz. It told all customers to assume every environment variable on the platform had been decrypted and to disable or rotate all keys. Days later it confirmed that multiple customers had reported unauthorized use of AI API keys they supplied to the platform.
An API key is cash on a billing account. Treat it that way.
The example
Picture a small shop that built a quoting helper with an AI tool. The owner pasted the model provider key into a hosting platform once, months ago, because that was the fastest way to get the demo running. It worked, so the key stayed there. Nobody wrote down where else that key lives.
Then the platform gets breached. The owner gets an email that says "rotate your keys." She changes the key on the platform dashboard and feels done. She is not done. The stolen key still works at the model provider until it is revoked there. The attacker does not need the platform anymore. The key is the access.
This is the part owners get wrong about keys. A password protects one account. A key spends money. And a key pasted into a platform you do not control is a copy you cannot take back. You can only kill it at the source and issue a new one.
Here is the honest counterpoint. Wiz itself says no single vulnerability or software dependency links the nine incidents, and no single attacker is confirmed. The root causes differ: credential access at some platforms, privilege escalation in build infrastructure at Railway, a third-party vendor dependency at Warp, a legacy system at OpenCode. Most investigations are still open and technical detail is limited. So do not read this as one gang running one campaign. Read it as nine separate warnings that stored keys are now a hunting ground, and the hunters move fast.
The next step
This week, make a list of every place you have pasted an AI API key: hosting platforms, developer tools, automation tools, and any dashboard that asked for one. For each key, do three things.
First, revoke it at the provider and issue a new one, even if the platform says it already rotated something on its side. The provider is the only kill switch that counts. Second, narrow the new key: limit it to the one model and the one job it needs, and set a spend cap or quota alert so a stolen key trips an alarm instead of running up a quiet bill. Third, write down where the new key lives, so the next breach notice is a ten-minute job instead of a scavenger hunt.
If you cannot list where your keys are, that list is the work. A key you cannot find is a key you cannot kill.
Want help finding the one bottleneck worth handing to AI first? A free 10-minute workflow conversation will map where the hours go and who signs off before anything final. No pitch, just the map.
Sources: Wiz, https://www.wiz.io/blog/black-october-for-ai-developer-platforms; BleepingComputer, https://www.bleepingcomputer.com/news/security/northflank-security-breach-leads-to-abuse-of-customers-ai-api-keys/; Northflank status, https://status.northflank.com. Opposing view: Wiz, https://www.wiz.io/blog/black-october-for-ai-developer-platforms.
Sources
The other side
Wiz itself says no single vulnerability or software dependency links the nine incidents, and no single attacker is confirmed. The reported root causes differ from platform to platform: credential access at some, privilege escalation in build infrastructure at Railway, a third-party vendor dependency at Warp, and a legacy system at OpenCode. Most investigations are still open and technical detail is limited, so this is a pattern of hunting for stored keys, not proof of one coordinated campaign.
Source: Wiz
Connected stories
"The AI agent didn't knock. It used the keys you left out."
Same topic: trust and verification; Same topic: cybersecurity
"One hacker breached 100 company websites for five grand in AI tokens"
Same topic: trust and verification; Same topic: cybersecurity
"Yelp's AI now answers your customers before you do"
Same topic: trust and verification
"The agent turned off its own alarm"
Same topic: trust and verification
Want this every week?
One idea, one example, one next step. Free, every Wednesday. Or step into the rooms where owners talk it through.