NVIDIA OpenShell CVE-2026-65086

Exploitation requires a low-privilege authenticated session with access to the NVIDIA OpenShell sandbox exec handler over the network (AV:N, PR:L). High attack complexity (AC:H) indicates that exploitation is not straightforward against default configurations - specific system state, timing, or precise input construction is likely required. The lack of user interaction (UI:N) means no victim action beyond the attacker's own authenticated session is needed once conditions are met. Exploitation is limited by: requirement for valid low-privilege credentials, non-trivial attack complexity likely involving precise exec handler input crafting, and deployment on Linux systems only per the product description. No special non-default feature flag is named in the description, but the AC:H metric implies conditions beyond basic network connectivity must be satisfied.

The CVSS base score of 6.8 (Medium) is consistent with the vector's AC:H and PR:L constraints, which meaningfully reduce exploitability relative to a trivially reachable unauthenticated injection. AV:N confirms network reachability, but AC:H signals that exploitation requires non-trivial preconditions - such as a race condition, specific system state, or precise crafting of exec handler input - making opportunistic mass exploitation unlikely. PR:L requires an authenticated low-privilege foothold, further narrowing the attacker pool. No EPSS score is available in the provided data, and neither KEV listing nor public POC has been identified, suggesting this has not been weaponized broadly at time of analysis. The absence of availability impact (A:N) in the CVSS vector is notable given 'data tampering' is listed as a consequence - this may reflect that the primary exploitable path affects confidentiality and integrity of data rather than service disruption. The wildcard CPE version range leaves affected version boundaries unconfirmed, which complicates patch prioritization.

An attacker with low-privileged network access to the NVIDIA OpenShell service crafts a malicious input payload targeting the sandbox exec handler, embedding shell metacharacters or command separators that are not properly sanitized. Under the right system conditions (AC:H), the injected commands escape the sandbox and execute in the context of the OpenShell process, potentially allowing the attacker to read sensitive files or modify data on the host. No public proof-of-concept exploit has been identified at time of analysis.