Actively exploited Security news
Barracuda details an ARTEX-style intrusion: a forgotten vendor key opened 10 cloud accounts
Security vendor Barracuda disclosed on 10 and 11 November 2026 that an attacker got into its cloud between 1 and 2 November using an access key it had once given an AI vendor and never revoked. The activity matches the ARTEX agentic framework. Barracuda found no direct evidence that customer credentials were viewed, but cannot rule it out and has contacted some customers.
Key points
- The attacker used a valid cloud access key that Barracuda had issued to a third-party AI vendor for an earlier project and never retired. The vendor's development environment was compromised in October, which is the likely source of the key.
- From 04:33 UTC on 1 November to 23:31 UTC on 2 November, the attacker chained the secrets it found into about 10 cloud accounts, a data analytics system, source code repositories and Kubernetes clusters, where it left a backdoor container.
- Barracuda found no direct evidence that the attacker viewed the credentials Barracuda XDR uses to collect customer data, but logging was not detailed enough to rule it out. It rotated them and contacted affected customers; if you have not heard from Barracuda, it says no action is needed.
- The tradecraft matches ARTEX. Barracuda stresses that this tells us how the attacker operated, not who it was, and compares the incident to the October breach at Railway.
What happened
Barracuda, which sells firewalls, email gateways and an XDR service, posted an advisory in its Trust Center on 10 November 2026 and a detailed write-up by its CISO, Arve Kjoelen, on 11 November. An attacker got into its cloud environment and, in roughly 44 hours, touched about 10 cloud accounts.
The way in was not a product flaw but a forgotten access key. Barracuda had given it to an outside AI vendor for an earlier project and did not retire it when the work ended. The vendor's development environment was breached in October, and Barracuda believes that is most likely where the attacker found the key.
How the attack unfolded
Within minutes the attacker had mapped what the key could reach and read stored secrets, including AWS access keys and Azure service principal secrets. Each secret became the next step: more cloud accounts, a data analytics system, a source code access token, then Kubernetes clusters.
On the second day, the attacker spent nearly four hours pushing short-lived branches to several repositories to trigger a build workflow that printed build-system secrets to the job log, encoded to slip past log masking. In Kubernetes it deployed a container that mounted the host file system read-only, and it tried to create a new administrative cloud identity, which was blocked. On some servers it went looking for cryptocurrency wallet material: Ethereum keystore files, seed phrases and extended private keys.
Barracuda bases the ARTEX match - the agentic AI penetration-testing framework attackers have been abusing since the summer (see our ARTEX report) - on the speed and sequence of actions and on the naming pattern of its probes. The company is explicit that this tells us how the attacker operated, not who it was. It compares the incident to October's breach at Railway.
Who is affected
One system within the attacker's reach stored the credentials Barracuda XDR uses to collect customer data. Barracuda found no direct evidence they were viewed, but logging there was not detailed enough to rule it out. It therefore treated everything in reach as compromised, rotated it, and contacted Managed XDR Cloud and Monitored Endpoint customers whose API integration credentials might be affected. Customers who have not heard from Barracuda need take no action, the company says. It does not report any compromise of its products.
How to check for compromise
Barracuda published no IP addresses, domains or hashes. It describes behaviour you can hunt for in your own environment:
- use of an old key issued to a vendor, especially from an unfamiliar address,
- probe names made of the letter
ifollowed by digits, with no readable words, - commands searching for the environment variables
WALLET,MNEMONIC,SEEDandPRIV, - short-lived branches that only trigger a build and print encoded text to the log (our Base64 decoder will unpack a suspicious string),
- a container that mounts the host file system.
What to do
- Customers Barracuda contacted: revoke and recreate your integration credentials following the steps in the company's email.
- Everyone: list the keys and tokens your vendors hold and revoke any nobody still needs.
- Replace long-lived keys with short-lived, narrowly scoped access, and make sure one key does not lead straight to more secrets.
- Turn on detailed logging for secret stores - without it, as this case shows, you cannot scope a breach after the fact.
- Assume an AI agent can do in a day what takes a human operator weeks. Build detections on behaviour, not just IP addresses.
Timeline
- October 2026 - the AI vendor's development environment is compromised.
- 1 Nov 2026, 04:33 UTC - first use of the key.
- 2 Nov 2026, 23:31 UTC - last recorded activity.
- 4 Nov 2026 - an engineer reports unusual account activity and the vendor gets in touch about its own incident; the backdoor container is removed and credential rotation begins.
- 10 Nov 2026 - Barracuda Trust Center advisory.
- 11 Nov 2026 - detailed write-up on Barracuda's blog.
Details come from Barracuda's write-up of 11 November 2026 (checked 16 November 2026). We did not separately review the Trust Center advisory and rely on the write-up's reference to it. Barracuda does not name the AI vendor.
FAQ
Were Barracuda products such as firewalls or email gateways compromised?
Barracuda does not say so. It describes an intrusion into its cloud environment and treats only integration credentials for Managed XDR Cloud and Monitored Endpoint as potentially exposed.
Do I need to do anything as a Barracuda customer?
Only if Barracuda contacted you - then revoke and recreate your integration credentials using its instructions. The company says customers who have not heard from it need take no action.
Who was behind the attack?
Barracuda has not attributed it. The ARTEX match describes the tooling and method, not the actor; security firms have seen both criminal and state-aligned groups using ARTEX.
What is the lesson for my own organisation?
Long-lived keys held by vendors are a weak point. Replace them with short-lived, narrowly scoped access, revoke them when a project ends, and make sure no single key leads straight to more secrets.
Sources
Spotted a mistake? Let us know and we will correct it openly.