Velociraptor Has a 9.9 Permission Bypass. Investigators Can Run Bash
- The Mess: Velociraptor has a CVSS 9.9 authorization flaw that lets users schedule sensitive artifacts through client monitoring while bypassing additional permission checks. One of those artifacts can execute arbitrary commands through Bash.
- The Damage: A low-privileged Velociraptor user can turn an investigation feature into a route for arbitrary command execution on monitored endpoints.
- The Fix: Upgrade Velociraptor to 0.77.2 or later and audit users who can schedule client-monitoring artifacts.
The Permission Check Was Missing
CVE-2026-19583 affects Velociraptor versions before 0.77.2.
The problem sits in client monitoring artifacts.
Velociraptor normally places additional permission requirements around dangerous artifacts. Linux.Sys.BashShell, for example, can execute arbitrary commands on endpoints, so it requires the EXECVE permission.
Client monitoring didn’t enforce that same check.
That is the hole.
A user who could schedule a client-monitoring artifact could use that mechanism to schedule artifacts that should have required additional privileges.
From Monitoring to Bash
This isn’t an ordinary information-disclosure bug.
The dangerous artifact is already built into the platform.
Linux.Sys.BashShell exists specifically to execute commands on endpoints. Under normal authorization rules, access to it is restricted.
CVE-2026-19583 provides a way around those restrictions.
The attack chain is therefore brutally simple:
Investigator → client monitoring → restricted artifact → Bash execution
No exploit chain involving memory corruption is required.
The application already has the dangerous capability.
The authorization layer simply failed to protect it.
The Second Check Was Missing Too
There was another problem in the same code path.
Client-monitoring artifacts were not required to carry the expected CLIENT_EVENTS type.
That meant the mechanism could be used to schedule artifacts that were never supposed to be available through client monitoring in the first place.
Two missing restrictions.
One very nasty result.
This Isn’t an Internet-Wide RCE
There is an important qualification.
CVE-2026-19583 requires the attacker to already have at least the Investigator role and the ability to add a client-monitoring artifact through the GUI.
So this isn’t:
Internet → Velociraptor → root
It’s:
compromised or malicious Velociraptor account → privilege bypass → command execution
That’s still serious.
Velociraptor is an endpoint visibility and incident-response platform. The people who can operate it already sit close to the machines being investigated.
Giving a lower-privileged account access to arbitrary command execution changes that trust boundary completely.
The Scope Can Change
The CVSS vector assigns the vulnerability Scope: Changed, with high confidentiality and integrity impact.
That makes sense.
The vulnerable authorization decision happens inside the Velociraptor server, but the abused artifact executes against monitored endpoints.
The security boundary isn’t limited to the management application.
The endpoint becomes the target.
Patch 0.77.2
Velociraptor lists 0.77.2 as the fixed version for CVE-2026-19583.
Administrators should also review who can:
- schedule client-monitoring artifacts
- access the Velociraptor GUI
- use investigator-level roles
- deploy or modify monitoring artifacts
- execute endpoint artifacts
If the system was exposed while vulnerable, review artifact scheduling and endpoint execution logs.
Look for unexpected Linux.Sys.BashShell activity.
Look for users invoking artifacts they normally shouldn’t need.
And don’t assume an Investigator account is harmless simply because it isn’t an administrator.
This Is Part of a Larger Problem
CVE-2026-19583 isn’t Velociraptor’s only recent authorization problem.
The project’s advisory list contains multiple 2026 issues involving permission bypasses, cross-organization access, identity spoofing, path restrictions and artifact execution.
That makes version management particularly important for security teams running Velociraptor as an internal investigation platform.
The tool has extraordinary visibility into endpoints.
Its authorization model needs to be treated as part of the security perimeter.
Bugstoday Opinion
This is the kind of bug that looks boring until you understand what the application can actually do.
There is no flashy packet.
No kernel exploit.
No browser sandbox escape.
Just a missing permission check.
But the feature behind that check can launch Bash on endpoints.
That’s the problem with security tooling: the more powerful the tool becomes, the more dangerous a small authorization mistake gets.
Velociraptor didn’t need another RCE.
It already had one.
It just needed someone to bypass the permission protecting it.
Today’s Bugs. Tomorrow’s Breaches.
Technical Sources
Velociraptor Security Advisory — CVE-2026-19583
CVE.org — CVE-2026-19583
Velocidex Velociraptor — Pull Request #4967
CIRCL Vulnerability-Lookup — CVE-2026-19583
