Bookmarks

    Nothing saved yet. Use the bookmark on any article. Bookmarks are kept on this device only: they stay private and work without an account.

    Settings

    Appearance

    Reading mode

    Font size

    Content width

    Line spacing

    Keyboard shortcuts

    /
    Search
    j k
    Next / previous story
    ← →
    Previous / next section (columns)
    b
    Bookmarks
    s
    Settings
    r
    Columns / scroll
    t
    Light / dark

    Saved in this browser only.

    Ping2Prompt
    Subscribe to the Weekly Brief Sign in
    A dark, faceted NVIDIA GPU workstation with a green NVIDIA logo sits on a wooden desk next to a monitor, keyboard and desk la
    Wikideas1, CC0 1.0, via Wikimedia Commons

    NVIDIA OpenShell for Linux: path allowlist bypass and other flaws disclosed

    Automated source check

    NVIDIA

    OpenShell path allowlist bypass is rated 8.4

    NVD has published a batch of CVEs for NVIDIA OpenShell for Linux, covering the L7 proxy and policy enforcement, sandbox SSH authentication and transport, Kubernetes sandbox configuration, API authorization and the inference-interception parser.

    The most severe is CVE-2026-100971 (HIGH, CVSS 8.4). In L7 path allowlist enforcement, a local low-privileged attacker can use semicolon-dot path sequences to traverse paths and get around allowlist restrictions. The listed impacts are information disclosure, data tampering and denial of service.

    Other OpenShell flaws, from high to medium severity

    • CVE-2026-100972 (HIGH, 8.1): missing resource-level authorization on create, read, update and delete operations lets an authenticated remote low-privileged attacker access or modify resources.
    • CVE-2026-100975 (HIGH, 7.6): on Kubernetes, the SSH handshake secret sits in a plaintext environment variable, so a local low-privileged attacker can read it and reuse it to reach other sandboxes.
    • CVE-2026-100974 (MEDIUM, 6.7): the SSH handshake secret is shared across sandboxes, so a remote attacker with high privileges can gain access to other sandboxes.
    • CVE-2026-100982 (MEDIUM, 6.5): the inference-interception parser has no body-size or chunk-count limits, so a sandboxed actor can exhaust resources with a chunked request.
    • CVE-2026-100970 (MEDIUM, 6.4): after a network policy hot-reload, existing L7-inspected tunnels keep using the old policy, which a local low-privileged attacker can abuse.
    • CVE-2026-100973 (MEDIUM, 6.4): nonstandard HTTP methods let a low-privileged attacker bypass L7 policy on HTTP CONNECT tunnels.
    • CVE-2026-100978 (MEDIUM, 6.4): the gateway-to-sandbox transport accepts any host key, so a machine-in-the-middle can impersonate the sandbox endpoint, intercept credentials and tamper with commands.
    • CVE-2026-100981 (MEDIUM, 6.4): the layer-7 REST proxy and the upstream server can read body framing differently, letting a sandboxed actor smuggle a second request past policy checks.

    A further Kubernetes flaw, CVE-2026-100976 (MEDIUM, 5.2), confuses TCP source-port identity so the proxy blames a different process, which lets a local low-privileged attacker bypass process-based network policy.

    Which OpenShell setups are exposed

    Most of the attacks need a foothold: a local low-privileged user, a sandboxed actor, or an authenticated account. If you run OpenShell sandboxes on Kubernetes, CVE-2026-100975 and CVE-2026-100976 apply to your setup.

    The sources do not list affected ranges or any other scores, and they do not say if the flaws are being exploited. According to NVD, the status of each record is "Received".

    No fixed OpenShell version is confirmed yet

    We found no fix in the sources, so the fixed-release table is empty. The NVD records for all of these CVEs reference NVIDIA's product-security entry 2026/5929 on GitHub. Check that entry for fixed releases and any mitigation, and update OpenShell as soon as a fixed build is listed.

    Sources

    Image: Wikideas1, CC0 1.0, via Wikimedia Commons


    This story was produced with AI assistance from the sources listed above and fact-checked against them before publication. Read our editorial methodology and the corrections log. Spotted an error? Email [email protected].

    Continue

    Share on Bluesky · Mastodon · Email ·

    Discussion