Canonical Juju CVE-2026-93336

The attacker must be a principal already authenticated to the Juju controller, but the required privilege is minimal: either a unit agent, or a human user holding only 'login' permission with no access to the target model (confirmed by the PR's TestUserWithoutModelAccessRefused test, which the fix now blocks). No special or non-default configuration is needed - this affects default deployments of Juju 4.0 through 4.0.15, and the attacker simply issues a request to the unit resources download route naming an arbitrary unit in any model. There is no user interaction requirement. The sole limiting factor is possession of a valid controller login-level credential or agent identity; a completely unauthenticated outsider cannot trigger it.

The CVSS 4.0 base score of 7.1 (High) is driven by a network-reachable (AV:N), low-complexity (AC:L) flaw needing no user interaction (UI:N) and only low privilege (PR:L), with high confidentiality impact (VC:H) and no scope change (SC/SI/SA:N). The decisive mitigating signal is PR:L: exploitation is NOT unauthenticated - the attacker must already be a controller-authenticated principal, but that bar is very low because even a user holding only 'login' permission (no model access) or any unit agent qualifies, as confirmed by the vendor advisory and the PR's authorization tests. No EPSS score was supplied, so statistical exploitation likelihood cannot be quantified; the CVE is not in CISA KEV, so there is no confirmed active exploitation. No weaponized public exploit is referenced - only the vendor advisory (GHSA-fp7x-6hq5-3c3w) and the fix PR (#23360), though that PR's diff and new tests effectively disclose the exact request mechanism (GET on the unit resources download route naming another unit). The high confidentiality impact is material because leaked artifacts include container-registry credentials, which can be reused to pull or push images outside Juju, so the practical blast radius exceeds pure read-only disclosure. There is no signal conflict here: all signals (High CVSS, vendor-reported, patch available, vuln.today 'this_week' priority) point to a genuine, prioritize-promptly issue rather than a high-score/low-real-risk finding, tempered only by the authentication prerequisite and absence of confirmed in-the-wild use.

An attacker who holds any controller credential - a compromised unit agent token, or a low-privilege user account with only login permission - sends a crafted GET request to the unit resources download route naming a unit belonging to another application or model. Because the route performs no cross-check tying the request to the authenticated entity, it returns that application's charm resource, including embedded container-registry credentials. No public weaponized exploit is referenced, but the fix PR (#23360) and its new authorization tests disclose the exact request path and parameters.