Hackers used a valid but unretired cloud access key and the ARTEX AI penetration-testing framework to breach Barracuda's cloud environment, touching roughly 10 cloud accounts and harvesting AWS access keys, Azure service principal secrets, and other credentials over approximately 44 hours1,2.
The activity spanned November 1–2, 2026. Barracuda disclosed the incident publicly on November 10.
How the key leaked
The initial access key belonged to a completed engagement with an unnamed AI vendor, but Barracuda had not retired it. A November 13 update confirmed that attackers obtained the key from a configuration file in the vendor's GitHub repository after compromising its development environment. Other Barracuda credentials in that file had expired or were subsequently revoked.
The attacker's first recorded request occurred at 04:33 UTC on November 1. Within minutes, it enumerated accessible resources and read stored secrets. Over the following day, stolen credentials enabled movement from cloud accounts into a data analytics system. The attacker then reached a source-code access token and Kubernetes clusters. Early on November 2, the attacker pushed short-lived repository branches to trigger build workflows that printed secrets into job logs. Encoding prevented conventional secret masking from concealing the values. Inside Kubernetes, the attacker deployed a container mounting the underlying server's filesystem, which, although restricted to read-only access, exposed host credentials. On some servers, the attackers gained shell access and searched for Ethereum keystore files, seed phrases, extended private keys, and other wallet-related information. Barracuda reported that no wallet material existed on those systems. An attempt to create an administrative cloud identity was blocked.
The malicious container was removed on November 4, Pacific time, though Barracuda said the attack left a backdoor behind in a Kubernetes cluster. Barracuda rotated secrets, restricted workloads, expanded logging, and began removing long-lived vendor keys. The company notified potentially affected customers to revoke and recreate integrations and review activity since November 1.
ARTEX's widening footprint
Barracuda linked the operational patterns to ARTEX, an AI-driven penetration-testing framework, but cautioned that identifying the tooling does not establish who operated it. Barracuda has not attributed the attack. The company said the attack was similar to the October intrusion at cloud application hosting platform Railway and fit a previously reported wave of intrusions at dozens of AI developer platforms.
Published on GitHub in July 2026, ARTEX relies on a planner to formulate an objective and create tasks, and on agents to perform those tasks, while the platform records progress, findings, and assets3. It reportedly used DeepSeek v4.1-flash as its initial model but can use other LLMs; GLM-5.3 and Grok 4.6 were observed in certain instances. Its users mainly rely on Chinese models, which SecurityWeek described as more permissive compared to US-made ones. ARTEX was designed for security testing, but threat actors have been using it in live intrusions since at least August, and its occurrence in real-world attacks has increased during recent weeks.
The FBI warned in a new alert that ARTEX "enables threat actors with minimal technical expertise to conduct complex, multi-target operations simultaneously, fundamentally accelerating attack velocity and expanding operational scale". The alert mentioned the framework's use in attacks on South Korean financial firms and ConnectWise, citing risks including faster attacks through automation, exploitation of neglected peripheral devices, and the ability to find and exploit supply chain attack vectors. Amazon said two ARTEX-powered campaigns relied on long-known techniques but operated at far greater speeds courtesy of AI, and that the agents succeeded where basic security hygiene was missing, including default or hardcoded credentials, keys left in code, unauthenticated interfaces, and unpatched software.
ANALYSIS The Barracuda breach underscores a concrete failure mode: a single unretired key in a vendor's repository gave an autonomous agent chain enough leverage to traverse cloud accounts, CI/CD pipelines, and Kubernetes clusters without exploiting any software vulnerability. Separately, a multi-agent AI exploit framework linked to a suspected Russian-speaking operator was reported the same week[1]. The FBI's ARTEX alert and Amazon's corroborating observations mark a second distinct agentic attack toolchain surfacing in the same period.