Kestra CVE-2026-45807

Exploitation requires an authenticated Kestra account with privileges to create or run at least one execution (CVSS PR:L), and the target instance must be running a vulnerable version (prior to 1.0.43 or 1.3.19) using the local storage backend that resolves kestra:// URIs through StorageInterface.parentTraversalGuard. The attacker must supply a URI whose traversal is URL-encoded (%2E%2E rather than literal '..') to bypass the guard. Limiting factors: a valid account and execution permission are mandatory (so unauthenticated internet attackers cannot exploit it directly), and the impact is bounded by the OS-level file permissions of the Kestra process account - files the process cannot read remain inaccessible. No user interaction or special client platform is needed.

The supplied CVSS:3.1 vector (AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:N/A:N = 7.7 High) indicates a network-reachable, low-complexity attack requiring only low privileges and no user interaction, with high confidentiality impact and a scope change - consistent with the description's claim that any authenticated user with one execution can read host files and cross tenant boundaries. The PR:L requirement is the main mitigating factor: the attacker must already hold a valid Kestra account with execution rights, which limits exposure on properly access-controlled instances but is significant in multi-tenant or shared deployments. No EPSS score, KEV listing, or POC indicator was provided, so likelihood of widespread automated exploitation cannot be quantified from the data; the absence of KEV/POC signals suggests no confirmed active exploitation, though the vulnerability is mechanically simple to weaponize once the guard bypass is known. The signals are internally consistent (no AV/PR conflict), so this is a credible real priority for any multi-tenant Kestra operator rather than a high-CVSS-but-low-real-risk issue.

An attacker who holds a low-privilege Kestra account creates or triggers an execution and submits a kestra:// URI containing a URL-encoded traversal such as kestra://...%2E%2E%2E%2E/etc/passwd; the guard passes the literal string but URI.getPath() decodes it, and the OS resolves the '..' to read the target file. The same technique reads mounted Kubernetes/Docker secrets or other tenants' execution outputs, exfiltrating credentials usable for further compromise. No public exploit was identified at time of analysis, but the technique is straightforward given the published advisory.