The software that is supposed to keep attackers off a Windows machine can, on unpatched systems, be talked into running their code instead. WatchGuard disclosed two critical vulnerabilities in WatchGuard Agent for Windows on 25 August 2026, and either one lets somebody who has never logged in execute code on the host.
The agent is the component of WatchGuard's endpoint protection that sits on each Windows machine, and it normally runs with SYSTEM privileges, the highest level Windows grants. That is what makes these bugs matter: the flaw and the privilege arrive together.
Two ways in, neither needs a password
CVE-2026-57909 carries a CVSS v4.0 score of 9.4 and is a path traversal flaw, meaning a file path that escapes the directory it was meant to stay inside. WatchGuard says a remote, unauthenticated attacker on an adjacent network can use it to run arbitrary code, and that successful exploitation results in complete loss of confidentiality, integrity and availability of the endpoint protection component. The advisory credits a finder listed as R31n.
CVE-2026-57910 scores 9.3 and is an improper authentication flaw. WatchGuard's write-up says an unauthenticated attacker with network access can abuse the agent's UDP discovery and command service, together with its TaskExecute event handler, to make the agent download and run a program of the attacker's choosing, typically with root or SYSTEM privileges. The vendor lists missing authentication for a critical function, improper verification of a cryptographic signature, and download of code without an integrity check among the underlying weaknesses.
WatchGuard says it is not aware of any exploitation of either flaw in the wild. Security agents are an attractive target precisely because of the privileges they hold, and ransomware crews already put real effort into blinding endpoint tools before they encrypt anything.
Everything below 1.25.13.0000 is affected
Both advisories list WatchGuard Agent for Windows as vulnerable at every version under 1.25.13.0000, and that release fixes both issues. For CVE-2026-57910, WatchGuard also names 1.17.02.0000 and 1.17.21.0000 as fixed versions, which matters for estates deliberately held on an older branch. There is no version of the agent that is safe from CVE-2026-57909 short of 1.25.13.0000.
Update to 1.25.13.0000, then firewall the UDP service
Peru's Centro Nacional de Seguridad Digital carried both flaws in its integrated digital security alert 159-2026 on 27 August, and the Peruvian guidance goes further than "install the update". The CNSD tells defenders to inventory every installation of the agent rather than assume the management console knows about all of them, to prioritise machines on shared network segments or segments reachable by third parties, to use firewall rules to restrict access to the agent's UDP discovery service and its command service, and to review logs for unauthorised downloads or executions.
That last pair is the useful part while patching is still in progress. Both flaws are reached over the network, so anything that narrows who can talk to the agent narrows the exposure. The full technical detail sits in WatchGuard's own advisories for CVE-2026-57909 and CVE-2026-57910, which the company published on 25 August and updated the following day.
Endpoint agents are installed on the machines an organisation cares most about, and they are trusted to run whatever their own update path hands them. A bug that lets a stranger on the network drive that path is worth treating as urgent even with no exploitation reported.
This briefing is provided by IntelFusions for informational and defensive purposes only. It is based on sources assessed to be reliable at the time of writing, and analytic judgments carry the confidence levels indicated. Indicators of compromise are defanged; re-arm them only in controlled environments. IntelFusions is not affiliated with the organizations named and makes no warranty as to completeness or accuracy.