Barracuda Networks says an attacker spent about 44 hours inside parts of its cloud environment on November 1 and 2, 2026, after logging in with a cloud access key the company had handed to an AI vendor and never retired. The probe names, the order of attacker activity and the automated hunt for credentials and crypto wallets match ARTEX, the open-source AI-driven penetration-testing framework. One of the systems reached stores the credentials Barracuda XDR uses to pull data from customers’ Microsoft 365, Google Workspace and endpoint tools.
Quick Take
- What: An attacker used a forgotten cloud access key, leaked from a compromised AI vendor, to move through about 10 Barracuda cloud accounts in roughly 44 hours.
- AI angle: Barracuda says the activity matches ARTEX, an AI agent framework where a “planner” agent picks the next step and “worker” agents run the tools.
- Who is affected: Only organizations using an API integration for Barracuda Managed XDR Cloud or Monitored Endpoint, and only those Barracuda notified directly.
- What to do: Notified customers should revoke and recreate their integration credentials and review activity in connected systems since November 1, 2026.
In plain English
A security company, Barracuda, was broken into because an old digital key it had given to a partner was never switched off. When that partner was hacked, the attacker found the key and used an AI-powered hacking tool to rush through Barracuda’s systems in under two days, looking for passwords and cryptocurrency. Barracuda found no direct sign that customer login details were viewed, but it warned customers who could be affected.
What you should do: If you are an ordinary user, you do not need to do anything. Only businesses that Barracuda contacted about their XDR connections need to act.
What happened
According to Barracuda’s technical write-up, the company had issued a cloud access key to a third-party vendor “working in AI” for an earlier project and did not retire it when the work ended. In October, that vendor’s development environment was compromised. Barracuda’s November 13 update adds that the key sat in a configuration file in the vendor’s GitHub repository.
The first call with the key came on November 1 at 04:33 UTC. Over the next 24 hours the attacker read stored secrets such as AWS access keys and Azure service principal secrets and used each one to reach the next system: from cloud accounts to a data analytics system, then to a source-code access token, then to Kubernetes clusters. Early on November 2 it pushed short-lived branches to several repositories to trigger a build workflow that printed the build system’s secrets into the job log, encoded so the usual log masking would not hide them. The last recorded activity was November 2 at 23:31 UTC.
The attacker also deployed a container on a Kubernetes cluster that mounted the host’s file system read-only to harvest that server’s credentials, and tried to create a new administrative cloud identity, which was blocked. Barracuda removed the container on the evening of November 4 (Pacific time) and rotated credentials across every system involved. On servers where it had shell access, the attacker searched for environment variables named WALLET, MNEMONIC, SEED and PRIV, Ethereum keystore files, extended private keys, seed phrases and wallet-related Kubernetes secrets. Barracuda says no wallet material existed there.
Why Barracuda points to ARTEX
The attacker named cloud and Kubernetes resources with the letter “i” followed by digits (for example i31911, i28324probe and i19engx1) and used GitHub workflow files named i19ghci.yml and i19ci2.yml. Barracuda says these labels, the sequence of activity and the built-in routines for hunting credentials and wallets are consistent with ARTEX, which was released in July 2026 for authorized security testing and has since been reported in use by criminal and state-aligned groups. Barracuda stresses that because many groups use ARTEX, the match shows how the attacker operated, not who it was. Earlier coverage of the framework: ARTEX AI hacking platform used in cybercrime campaigns and the FBI FLASH on ARTEX intrusions reaching privileged access in under 3 hours.
What you should do
- Notified Barracuda customers: revoke and recreate integration credentials for Managed XDR Cloud or Monitored Endpoint, and review connected system activity logs from November 1, 2026 onward. Barracuda points to guidance on Barracuda Campus and to SOC@Barracuda.com for questions.
- Not notified: Barracuda says no action is required. Organizations that use neither integration are not affected.
- Threat hunting: Barracuda lists indicators including the IP addresses 180.178.56.10, 52.128.238.126 and 221.146.68.107, and user agents such as “Mozilla/5.0 (compatible; read-only-audit)” and Boto3/1.43.122. Their presence alone does not confirm malicious activity.
- Everyone running cloud infrastructure: expire keys issued to vendors when an engagement ends, remove long-lived and over-privileged access keys, and limit which workloads can read secrets. These are the fixes Barracuda says it is now making itself.
What it means
The key lesson Barracuda draws is speed: by the time an engineer reported unusual activity on November 4, the attacker had already finished its run in less than two days. Agentic tooling turns every leaked secret into the next hop within minutes, so detection that assumes human-paced intruders arrives too late. Barracuda also notes that some secrets within reach were credentials for AI services and that a motive around AI compute “cannot be ruled out.” It compares the chain to the October Railway incident, covered in our report on how the Railway breach exposed AI provider keys and led to $740,000 in unauthorized charges. More incidents are tracked in our AI Security hub.
Sources
- Barracuda Trust Center advisory “Barracuda XDR Integration Credentials” (published November 10, 2026, updated November 13, 2026)
- Barracuda blog, “Forty-four hours: Anatomy of a machine-speed intrusion” by Arve Kjoelen (November 11, 2026)



