Wiz links nine AI platform incidents to stolen credentials and model-key abuse
At least nine AI developer platforms disclosed security incidents in October, with public reports describing targeted searches for stored environment variables, especially those granting access to AI models. Wiz says it found two enterprise keys compromised through vendor breaches later used for roughly 800,000 and 1.7 million Bedrock calls; it estimates the usage in those cases alone could cost over €92,000 (over $100,000). The incidents remain under investigation, and Wiz says no single vulnerability or dependency links them.
Bottom line — Wiz’s two cases show stolen model keys can fuel millions of calls and more than €92,000 in estimated usage.
Go deeper 6
-
Wiz says both stolen keys were first tested within 30 minutes of credential access reported in a vendor breach.
-
In one case, around 90% of 800,000 Bedrock calls succeeded; in the other, more than a million of 1.7 million calls succeeded.
-
Wiz attributes the different throttling levels partly to key permissions: one allowed global cross-region routing, while the other was limited to US-region profiles.
-
Wiz traced the keys to infrastructure associated with discounted LLM resale sites, but says it cannot definitively link those operators to the initial breaches or the high-volume abuse.
-
Wiz reports that Fireworks, Zeabur and Lightning cited credential access, while Railway pointed to privilege escalation in build infrastructure and Warp to a third-party dependency.
-
Separate reporting by Akamai, SonarSource and Vorlon describes Mini Shai-Hulud supply-chain attacks that harvested developer credentials and added persistence in AI coding-tool configurations.