A heap-based buffer overflow vulnerability exists in Autodesk AutoCAD when parsing maliciously crafted DWF files. Exploitation can lead to application crashes or arbitrary code execution within the context of the current process. The vulnerability requires local access with user interaction. It has a high severity score of 7.8 CVSS v3.1.
Atomic Server is an open source headless CMS and real-time database. Prior to 0.40.3, unauthenticated requests to /bookmark can supply a url that server/src/plugins/bookmark.rs passes through fetch_data to fetch_body. Endpoint dispatch in lib/src/db.rs invokes the handler before check_read, and lib/src/client/helpers.rs previously checked only that the URL began with HTTP, without restricting loopback, private, link-local, DNS-rebinding, or redirect targets. The handler reflects fetched content in returned resource fields, making this a full-response server-side request forgery that can expose internal services and plain-GET cloud metadata. The short request timeout limits slow targets, and metadata services requiring a separate token request are outside the demonstrated GET-only path. This issue is fixed in version 0.40.3.
qbee-agent contains a local privilege escalation vulnerability due to improper handling of symbolic links and ownership management in its file manager logic. This flaw allows unprivileged users to escalate to root by exploiting symlink following in SSH key management, overwrite critical system files during RAUC updates, or change ownership of sensitive files via file distribution. The issue affects versions prior to 0.0.0-20260615134032-0fdf8e411de5. A patch is available in version 0.0.0-20260615134032-0fdf8e411de5 and later. Mitigations include disabling SSH key management for untrusted users, enforcing strict permissions on user home directories, and enabling kernel symlink protections.
Dell SupportAssist OS Recovery, versions prior to 5.5.16.3, contain an Untrusted Search Path vulnerability. A low privileged attacker with local access could potentially exploit this vulnerability, leading to Elevation of privilege.
GPAC before 26.07.0 contains a stack-based buffer overflow vulnerability in gf_filter_pid_resolve_file_template_ex() in src/filter_core/filter_pid.c that allows local attackers to write past fixed GF_MAX_PATH stack buffers. Attackers can convince a user to segment a crafted NHML input referencing an overly long media file name with MP4Box -dash to crash the process.
GPAC before 26.07.0 contains a type confusion vulnerability in the XMT scene loader that allows local attackers to corrupt memory by supplying unbalanced command buffer closing elements. Attackers can convince a user to open a crafted XMT-A file with MP4Box, causing xmt_node_end() to treat a GF_Command as an SFCommandBuffer and crash the process.
fzf is a command-line fuzzy finder. Prior to 0.74.5, trigger-driven zsh completion in shell/completion.zsh uses _fzf_feed_fifo to create a FIFO at the predictable shared path ${TMPDIR:-/tmp}/fzf-complete-fifo-$$ and continues after removal or creation errors. On a multi-user host, a local attacker can pre-create that path as a symlink so the victim's completion stream overwrites a victim-writable file, potentially enabling code execution when that file is later sourced. A pre-created FIFO can instead disclose a bounded stream of completion candidates or block completion. On default modern hosts, fs.protected_symlinks blocks the overwrite variant, and protected_fifos and protected_regular further limit the FIFO disclosure and blocking variants, so exploitation depends on the relevant kernel protections being disabled and on the victim invoking the affected completion. This issue is fixed in version 0.74.5.
ZoneMinder is a free, open source closed-circuit television software application. Prior to 1.38.4, detaintPath() in web/includes/functions.php can be bypassed with a PHP stream-wrapper path when sanitizing modal.php input. A remote unauthenticated attacker can submit such a path through the affected input and potentially read unauthorized local files through local file inclusion. This issue is fixed in version 1.38.4.
NVIDIA OpenShell for Linux has a vulnerability in its network policy hot-reload component. A local attacker with low privileges can bypass updated network policy restrictions by exploiting existing L7-inspected tunnels that continue to enforce the old policy after a reload. This flaw could lead to information disclosure and data tampering.
Teable is an AI spreadsheet for business. Prior to release.2026-07-21T11-27-40Z.2320, a registered user can supply path traversal sequences in the hash field accepted by POST /api/attachments/signature when the default local storage provider is in use. In apps/nestjs-backend/src/features/attachments/plugins/local.ts, LocalStorage.presigned uses the client-controlled hash as the storage filename, and LocalStorage.save resolves that path and copies the uploaded request body without calling assertPathWithinStorage on the write path. The read path uses the jail, but the write path does not, allowing the resolved destination to escape the configured storage directory. Because the backend container runs as uid 0, the attacker can write or overwrite arbitrary root-owned files, which can corrupt the instance or lead to root code execution. Object-storage deployments that do not use LocalStorage.save are not affected by this sink. This issue is fixed in release.2026-07-21T11-27-40Z.2320.
NVIDIA OpenShell for Linux contains a vulnerability in L7 REST policy enforcement where a local attacker with low privileges could exploit inconsistent HTTP interpretation by using cleartext HTTP/2 (h2c) over an HTTP CONNECT tunnel to bypass policy restrictions. A successful exploit of this vulnerability might lead to information disclosure and data tampering.
NVIDIA OpenShell for Linux contains a vulnerability in Kubernetes network policy enforcement where a local attacker with low privileges could cause the proxy to attribute network traffic to a different process due to TCP source-port identity confusion, allowing the attacker to bypass process-based network policy restrictions. A successful exploit of this vulnerability might lead to information disclosure and data tampering.
NVIDIA OpenShell for Linux contains a vulnerability in L7 policy enforcement for HTTP CONNECT tunnels where a local attacker with low privileges could cause a protection mechanism failure by using nonstandard HTTP methods to bypass policy restrictions. A successful exploit of this vulnerability might lead to information disclosure and data tampering.
NVIDIA OpenShell for Linux contains a vulnerability in the L7 path allowlist enforcement component where a local attacker with low privileges could cause path traversal by using semicolon-dot path sequences to bypass allowlist restrictions. A successful exploit of this vulnerability might lead to information disclosure, data tampering, and denial of service.
Eclipse Zenoh's zenoh-pico implementation, versions 0.7.2-rc up to but excluding 1.10.2, does not validate the Zenoh identifier (ZID) length when deserializing timestamps. An attacker-controlled length greater than 16 causes an out-of-bounds write to a fixed 16-byte identifier buffer and causes the subsequent zero-fill operation to receive a negative length converted to a large unsigned size. Truncated timestamp inputs can also cause out-of-bounds reads or assertion failures. A remote attacker able to deliver crafted timestamp-bearing Zenoh messages through an accepted transport can cause memory corruption and denial of service, with potential remote code execution in builds lacking appropriate binary protections. This issue is fixed in version 1.10.2.
Depending on the release and message-decoding path, the unchecked timestamp identifier length causes a heap-based or stack-based buffer overflow. Ordinary-frame network message decoding moved from heap-backed storage to stack-local storage in version 1.5.0. Earlier affected releases already contained stack-based decoding paths, including reassembled fragmented messages.
MLflow versions from 3.10.0 up to but not including 3.17.0 contain a vulnerability that allows unauthenticated remote attackers to read arbitrary files on the server. This occurs because the prompt branch in the _create_model_version() function accepts schemeless relative sources without proper path validation. The flaw enables attackers to specify server-local paths as model version artifact locations and retrieve files via the /api/2.0/mlflow/model-versions/get-artifact endpoint. Default unauthenticated servers are fully exposed, while basic-auth deployments are vulnerable to authenticated users with model creation privileges. The issue is fixed in version 3.17.0.