Hackers Exploit Exposed AI Vendor Key to Breach 10 Barracuda Cloud Accounts in 44 Hours

Blog WriterCybersecurity News - Original News Source is cybersecuritynews.com


Hackers used a cloud access key exposed through an unnamed AI vendor to enter Barracuda’s cloud environment and reach about 10 accounts in roughly 44 hours. The November 1–2 intrusion crossed cloud services, data analytics, source code, and Kubernetes systems as the attacker collected credentials to keep moving.

Barracuda said the key remained active after the vendor finished its project. Its security advisory, updated November 13, confirmed that the credential was stored in a configuration file inside the vendor’s GitHub repository.

Attackers obtained it after compromising the vendor’s development environment in October. Other Barracuda credentials in that file had expired or were later revoked.

According to Barracuda’s account, the first recorded request arrived at 04:33 UTC on November 1. Within minutes, the attacker checked what the key could access and read stored secrets, including AWS access keys and Azure service principal secrets.

Over the next day, those credentials gave the attacker access to a data analytics system, then a source code token, and finally Kubernetes clusters. Each newly found secret became another route into the environment.

Early on November 2, the attacker pushed temporary branches to several repositories to trigger build workflows. These workflows printed build secrets into job logs.

Encoding the values helped bypass controls designed to hide secrets in those logs. The technique echoes credential theft through malicious GitHub Actions workflows, where trusted build systems become tools for stealing cloud and AI keys.

Barracuda Cloud Account Breach

Barracuda linked the activity to ARTEX, an open-source AI-driven penetration testing framework, based on matching resource names, workflow patterns, and automated credential searches, though this does not identify the attacker.

The attacker searched infrastructure state files, server metadata services, and Kubernetes service accounts for credentials. Commands also searched for WALLET, MNEMONIC, SEED, and PRIV environment variables, Ethereum keystores, private keys, and wallet-related Kubernetes secrets. Barracuda said no wallet material existed on those systems.

A container deployed inside one Kubernetes cluster mounted the underlying server’s file system. Although access was read-only, it allowed credential harvesting. An attempt to create a new administrative cloud identity was blocked.

Barracuda traced the unusual account activity reported on November 4, removed the affected container that evening, and rotated credentials across affected systems.

One reached system stored integration credentials for Barracuda XDR. Investigators found no direct evidence that these credentials were accessed, but incomplete logging prevented them from ruling it out.

Notified customers must recreate integration credentials and review connected systems’ activity from November 1. Customers not notified need no action under the advisory.

The incident follows the Railway AI credential breach, where stolen keys generated about $740,000 in unauthorized charges. Railway’s report describes attackers targeting AI provider credentials for possible resale.

Barracuda has not confirmed that same goal, though AI service credentials were within reach. Barracuda is removing long-lived keys, improving audit logs, and tightening Kubernetes controls to limit future credential theft and reduce cross-system access.

Give your SOC the alert context on the latest threats to stop incidents quickly -> Integrate Threat Intel from 16K SOCs

The post Hackers Exploit Exposed AI Vendor Key to Breach 10 Barracuda Cloud Accounts in 44 Hours appeared first on Cyber Security News.