0
log entries a successful bypass leaves in Xen Orchestra, per Vates
2
advisories in the chain the attacker used; the second has no fix yet
8
IP addresses on the Vates blocklist, plus one Tor hidden service
10 Nov
the day Vates published both advisories, in 2026
The VM console is the estate
The management console for your virtual machines is not one server among many. It is the server that can start, stop, copy and delete every other one, and it holds the passwords to the place your backups live. When an attacker owns it, they do not need a foothold on each workload. They have the switch for all of them.
That is the lens for VSA-2026-062. Vates, the company behind XCP-ng and Xen Orchestra, says a threat actor is exploiting an authentication bypass in the console right now, that a compromise can end in root on every managed host, and that the bypass itself leaves no log. The fix for the login bug shipped on 10 Nov 2026. The second flaw the attacker used to get from the console down to the hosts, published the same day as VSA-2026-063, has no fix yet.
So the patch is necessary and it is not the end of the job. The questions that decide the outcome are older than this bug: who can reach the console, who inside it may import a virtual machine, and whether you could tell, after the fact, that someone had been there. On internal tests the virtualization console is one of the first things we look for, and it is often reachable from networks it was never meant to serve. That is the condition this advisory punishes.
What is VSA-2026-062 and who is affected
Xen Orchestra is the web console Vates ships for managing XCP-ng pools, sold as a packaged virtual machine, the Xen Orchestra Appliance, and also built from source by many smaller teams. VSA-2026-062, published on 10 Nov 2026, rates the flaw Critical on the Vates scale. There is no CVSS score yet; the field reads "Not available yet".
The root cause, in the vendor's own description, is a type confusion in the REST API's cookie handling. The authentication cookie is parsed in a way that can turn it into an empty object instead of a string. The REST authentication code passes that object to the token provider without checking that it is a string, the token store treats the empty object as an unfiltered lookup and returns the first token it holds, and the request is authenticated as that token's owner, provided the token has not expired. One more authenticated call then returns reusable login token IDs. Vates states the result plainly: "An adversary can bypass authentication and perform actions with 'Admin' privilege."
The advisory states active exploitation, not a proof of concept. It names the infrastructure the actor operated from initially, one IP address and one Tor hidden service, and its blocklist runs to eight IP addresses in all, plus that hidden service. The Hacker News reports that it is unclear when the activity started and whether the flaw was exploited as a zero-day. It also reports that the initial IP is flagged on VirusTotal as a RedLine Stealer command server. That is a reputation signal on an address, not an attribution, and Vates makes no attribution at all.
-
Product //
Xen Orchestra and the Xen Orchestra Appliance -
Flaw //
REST authentication bypass to Admin privilege -
Affected //
before 6.9.2 (stable) and 6.10.1 (latest) -
Fixed packages //
xo-server 5.211.2 (stable), 5.213.1 (latest) -
Severity //
Critical (Vates); no CVSS score and no CVE yet -
Exploited //
Yes, per Vates -
Logged //
No log of the bypass, per Vates
One practical warning on versions. Vates gives the fixed releases as 6.9.2 and 6.10.1. The Hacker News reported them as 6.9.1 and 6.10.0. Both name the same xo-server packages, 5.211.2 and 5.213.1, so check the package version on the running server rather than the release label on a dashboard. For source builds, Vates says to update to the latest master.
How a login cookie becomes root on the host
The two advisories read together describe the route. VSA-2026-063 says its import path "has been abused by a threat actor" using the Admin privilege gained through VSA-2026-062. Here is that route at the level of what each stage gives the attacker.
Figure 01
From one cookie to every host
- 01
In //
A malformed cookie authenticates as a token owner.
- 02
Admin //
Reusable login tokens with Admin rights.
- 03
Import //
A crafted VM import reaches the host.
- 04
Root //
Root on the XCP-ng host.
- 05
Stay //
Scheduled tasks and hidden files on the host.
Result //
The pool, its guests and its backup credentials.
Stage one costs the attacker nothing but network reach to the console. No password, no phishing, no second factor: the request is simply treated as someone who already logged in. Stage two turns a single request into durable access, because the tokens it returns keep working.
Stage three is the step that matters most for what comes next. A virtual machine image carries configuration as well as a disk, and VSA-2026-063 says certain configuration keys in an imported image were not filtered. Vates puts the consequence in one line: a user with import rights "could import a malicious VM (XVA) using a crafted keys parameter to compromise an XCP-ng host". The host is the physical server, the layer below every guest on it.
Stage five is what Vates found on compromised hosts. The advisory's indicators include cron entries under /etc/cron.d named zz-ustar and zz-ustar2, hidden files at the root of the filesystem named .ustar5 and .ustar6, and a running VM named pivot2. Its check command also lists /etc/ld.so.preload, a file that should not exist on most systems. That is the shape of an operator settling in, not a smash and grab.
At the end of the chain the attacker holds the hypervisor, the guests on it, and the configuration of the console, which includes the credentials for every backup remote. Vates' own rotation list makes the scope clear: proxy credentials, backup remotes on NFS, S3, Azure and SMB, the root account and pool secret, user accounts, OTP secrets and mail server passwords.
Why the patch is only half the answer
Three facts in the advisories keep this from being a patch and move on story.
The first is VSA-2026-063 itself. Its status is "In progress", it lists Xen Orchestra 6.10.1 and earlier as affected, and it says neither the latest nor the stable channel includes a fix. The only mitigation offered is to allow VM import to Admin users only and disable any account holding the import permissions over the JSON-RPC or REST API. Press coverage disagrees on who can reach it: SecurityWeek wrote that the zero-day requires administrator privileges, while the Vates advisory says an unprivileged user holding the import permission could use it. Take the vendor's wording. Any delegated operator account with import rights is, today, an account that can become root on the host.
The second is the logging. Vates is unusually direct: "In the case of a successful exploitation, there is no log to detect unauthenticated access." The access log that most teams would open first will look clean. Vates' detection advice is network capture of HTTP requests, and host checks for the indicators above. It also warns that those indicators "only cover this specific campaign" and that their absence "is not sufficient to conclude that no compromise has occurred." For a confirmed compromise it recommends a full reinstallation, because cleanup "cannot guarantee the removal of all backdoors".
A clean log here is not evidence of anything, because the bug never writes one. If your console was reachable by anyone you do not trust in the weeks before 10 Nov 2026, treat it as a host review, not a log review.
The third is the missing identifier. Neither flaw has a CVE ID. SecurityWeek and The Hacker News both say so, and when we checked the CISA Known Exploited Vulnerabilities feed on 12 Nov 2026, catalog version 2026.11.10, it had no Xen Orchestra entry. A vulnerability program that keys on CVE IDs or on the KEV list has nothing to match against this campaign. We made the same point about the OpenSSH bypass on Alpine that shipped with no CVE: the feeds lag the vendors, and an exploited flaw on your management plane will not wait for a number.
None of this is unique to Vates. We wrote recently about VeloCloud Orchestrator, another console that sits above everything it manages. Management planes concentrate power by design. That is their job. The control that matters is how few people and networks can reach them.
How to detect Xen Orchestra exploitation
Start where Vates points. The bypass shows in the HTTP request, in the authentication cookie, so the place to catch it is a reverse proxy, a web application firewall or a network sensor in front of the console. A legitimate session token is a plain string. The malicious requests carry a JSON-encoded value in that cookie instead. If you already terminate TLS for Xen Orchestra on a proxy that logs cookies, the rule below is a starting point.
Figure 02
Xen Orchestra cookie carrying a JSON value
Sigma
title: Xen Orchestra authentication cookie with a JSON encoded value
status: experimental
description: Requests whose authenticationToken cookie is JSON encoded, the pattern Vates lists for VSA-2026-062
references:
- https://docs.vates.tech/security-advisories/advisories/2026/vates-sa-2026-062/
logsource:
category: proxy
detection:
selection:
cs-cookie|contains: 'authenticationToken=j:'
condition: selection
falsepositives:
- Should be rare; a normal session token is a plain string
level: high Two cautions. Vates warns that a WAF rule on this pattern "can be bypassed using other formats", so use the rule to find past attempts and to alert, and use the upgrade to fix. And many deployments never log cookies at all; if yours does not, the rule has nothing to read and you are left with the host checks.
On the hosts, the advisory's checks are short. List VMs whose platform settings carry the hvm_serial file parameter, look for a VM named pivot2 among running VMs and in the XCP-ng xensource logs, and look for the zz-ustar cron entries, the .ustar files at the filesystem root and an /etc/ld.so.preload file. Any hit is a confirmed compromise in Vates' terms, and the guidance is to rotate every secret the console held and reinstall.
Then look outward. Block the eight IP addresses on the advisory's list at ingress and egress, block Tor from the management network, and search firewall logs for any past connection to them from the console or the hosts. A management network that can open connections to Tor is a finding on its own.
How we test a virtualization console on an engagement
We do not need this bug to ask the question it asks. On an internal test, finding the virtualization console is part of the first hour, and our internal network first hour guide covers the discovery. What we record is who can reach it: the user VLAN, the VPN pool, the guest wireless, a jump host shared with contractors. Every network on that list is a network from which an unauthenticated bug in the console is a pool takeover.
From the outside, the question is whether the console answers on the internet at all. Vates' first mitigation is "Do not expose Xen Orchestra to the Internet", and that is a fact an external test settles in minutes. Our exposed admin and debug endpoint checks show how we identify an admin page from its responses without logging in, and atarus-recon on our open source page covers the wider external inventory.
Inside the console, with credentials the client issues for the test, we map delegation. Which accounts hold import rights, which hold network creation rights, and which are shared. VSA-2026-063 turns that list from a hygiene note into the list of accounts that can reach root on the hosts. We also read where backup credentials live and whether the backup target accepts deletion from the same credentials the console uses, because that is the difference between a bad week and a lost estate.
Then we check what the client could see. Does a proxy log the requests to the console? Do the hosts ship their logs anywhere the console cannot reach? Is there an alert when a VM is imported or a host gets a new cron entry? We report how many steps separate a user laptop from root on the hypervisor, and which single change removes the most of them. Usually that change is a firewall rule, not a patch.
What is still coming
Expect the second fix and plan for the gap. VSA-2026-063 is marked in progress. Until it ships, the import restriction is the control, and it costs some convenience for delegated operators. Pay it.
Expect wider exploitation. The advisory now describes the root cause in enough detail for others to rebuild it, and Vates itself says exploitation by other actors with different infrastructure "cannot be ruled out". Consoles still exposed to the internet after this week are the obvious audience.
Expect identifiers to follow. A CVE ID, a CVSS score and possibly a KEV listing may arrive later, and vulnerability scanners will begin to flag the version afterwards. None of that changes what to do today, and waiting for it is how an exploited flaw on a management plane sits unpatched for a month.
The class will outlive this advisory. Web consoles that manage hypervisors, backup platforms and network orchestrators keep shipping REST APIs alongside older interfaces, and the newer path does not always inherit the older path's checks. The defence is the same in every case: reach limited to a management network, delegation kept narrow, and logs kept somewhere the console cannot touch.
What to do today
-
Upgrade and confirm the running package. xo-server 5.211.2 on the stable channel or 5.213.1 on latest; source builds to the latest master. Check the package on the server, not the release label, since reports disagree on the version numbers.
-
Take Xen Orchestra off every network you do not fully trust. Put it on a dedicated management network, block it from the internet, and block Tor and the advisory's eight IP addresses at ingress and egress.
-
Restrict VM import to Admin users until VSA-2026-063 is fixed. Disable accounts holding the import or network creation permissions over JSON-RPC or REST, as Vates advises.
-
Run the host checks on every XCP-ng host. The hvm_serial file setting, a VM named pivot2, the zz-ustar cron entries, the .ustar files and /etc/ld.so.preload. On any hit, rotate every secret the console held and reinstall.
-
Look for the cookie pattern in any proxy or WAF logs you have. The Sigma rule above is a starting point; keep the alert after you patch.
-
Track this flaw by its VSA number, not a CVE. Add VSA-2026-062 and VSA-2026-063 to your vulnerability tracker by hand so the second fix is not missed when it ships.
SOURCES
- Vates VSA-2026-062 // the Critical authentication bypass, active exploitation, no log of the bypass, affected and fixed versions, indicators and mitigations
- Vates VSA-2026-063 // the VM import flaw the attacker abused with Admin privilege, affected through 6.10.1, no fix yet, import limited to admins as the mitigation
- SecurityWeek // independent report of both flaws exploited, no CVE identifiers assigned, unauthorized access not logged
- The Hacker News // no CVE yet, unclear when exploitation began or whether it was a zero-day
- CISA KEV catalog // no Xen Orchestra entry in catalog version 2026.11.10, checked 12 Nov 2026