hydra-ray-launcher CVE-2026-115749

Exploitation requires the attacker to control configuration values or command-line overrides consumed by a Hydra application that has hydra-ray-launcher installed, and to supply a ${ray_pkg_version:<module>} interpolation; the resolver is registered at plugin discovery even when a different launcher is selected, so simply choosing another launcher does not prevent it. Harm is further gated on a target module with a damaging import-time side effect being already installed and importable - the resolver does not install attacker code and does not pass arbitrary function arguments, and AWS credentials and Ray job execution are not prerequisites. Limiting factors: the config/CLI-control prerequisite (PR:L), local attack vector (AV:L), the need for an existing harmful-on-import module, and that affected versions are only the pre-release/legacy ranges listed; Hydra 1.4's target-execution whitelist does not mediate this path and Hydra 1.3's blocklist is only defense-in-depth.

Signals point to a moderate, bounded issue rather than an urgent priority. The CVSS 4.0 vector (AV:L/AC:L/AT:N/PR:L/UI:N) scores 4.8 (medium) and the only impact is VI:L - integrity to the vulnerable system is Low, with VC:N, VA:N and no subsequent-system impact, matching the advisory's statement that the resolver neither installs attacker code nor passes arbitrary function arguments. PR:L and AV:L mean the attacker must already be able to influence the application's configuration or CLI overrides locally, which substantially narrows real-world exposure. No EPSS score is present in the input, the CVE is not in CISA KEV, and no PoC/exploit reference is supplied - so there is no public exploit identified at time of analysis and no evidence of active exploitation. The vuln.today Decision label is 'this_month' (not act_now/emergency), consistent with the limited impact. Harmful outcome is conditional: it requires that a module with a damaging import-time side effect already be installed and importable in the process. Net: a legitimate but low-to-moderate priority supply-chain/config-trust hardening fix, not a high-risk remote-code-execution emergency. Missing EPSS data means widespread-exploitation probability cannot be quantified from the input.

An attacker who can influence a Hydra application's configuration files or command-line overrides supplies a ${ray_pkg_version:<module>} interpolation naming an installed module chosen for its import-time behavior. Because plugin discovery registers the ray_pkg_version resolver even when another launcher is selected, the resolver calls importlib.import_module() on the attacker-named module, executing its import-time side effects in the application process. No public exploit code is identified, and impact is limited to side effects of an already-installed module (no attacker code is installed and no function arguments are passed).