CVE Watch14 sept 2026

CVE-2026-55832

Tract is a tiny, no-nonsense, self-contained TensorFlow and ONNX inference toolkit. Prior to 0.21.17, 0.22.3, and 0.23.2, the tract-onnx cra

CVSS

6.1

Medio

EPSS

—

KEV

—

Exploit Today

—

0-100

Publicado: 14 sept 2026 · Última mod.: 14 sept 2026 · CWE-22

EPSS · 30d

Sin historial EPSS suficiente todavía.

Descripción técnica

Tract is a tiny, no-nonsense, self-contained TensorFlow and ONNX inference toolkit. Prior to 0.21.17, 0.22.3, and 0.23.2, the tract-onnx crate passes the attacker-controlled external_data location from an ONNX model through onnx/src/tensor.rs get_external_resources and joins the value to the model directory without rejecting absolute paths or parent directory components. Loading an untrusted model through model_for_path can therefore make onnx/src/data_resolver.rs MmapDataResolver open an arbitrary local file and place the file contents into model tensors or inference output. Attacker-controlled offset and length fields can also select an out-of-range mapping slice and cause a denial of service, but the flaw does not write files or execute code. This issue is fixed in versions 0.21.17, 0.22.3, and 0.23.2.

Referencias oficiales

CVEs relacionados

CVECVSSEPSSKEVRExplotTítuloVis.

CVE-2026-558466.2 MED

—

———Allure 2 is the version 2.x branch of Allure Report, a multi-language test reporting tool. Prior to 2.39.0, the HTTP server started by allure serve and allure open uses URI.getPath() in Commands.setUpServer() in allure-commandline/src/main/java/io/qameta/allure/Commands.java and passes the percent-decoded request path to reportDirectory.resolve() without normalizing the result or confirming that it remains inside that directory. An unauthenticated client that can reach the server can submit parent-directory segments, including percent-encoded segments, and cause serveFile() to return any regular file readable by the Allure process. The server binds to localhost by default, but the --host option can expose it to other systems, and local users, adjacent containers, or browser-origin attacks may reach a local listener. This can disclose credentials, configuration, source code, build secrets, and other CI/CD data. This issue is fixed in version 2.39.0.5h CVE-2026-541788.1 ALT

—

———backpack/crud provides Create, Read, Update & Delete (CRUD) functions for Backpack, a collection of Laravel packages that help users build custom administration panels. Prior to 6.8.12 and 7.0.35, HasUploadFields::uploadMultipleFilesToDisk in src/app/Models/Traits/HasUploadFields.php trusts disk-relative paths from clear_<attribute>[] and passes them to Storage::disk()->delete without confirming that the paths are persisted on the current model record. An authenticated user with Update access to a CRUD using this mutator through src/app/Models/Traits/CrudTrait.php can delete another record's attachment, a shared asset, or another operational file on the configured disk by submitting its path. The newer MultipleFiles uploader is not affected because it intersects requested deletions with the record's persisted file list. This flaw does not permit reading the deleted files. The 5.x line remains affected through its final releases. This issue is fixed in versions 6.8.12 and 7.0.35.5h CVE-2026-54150—

—

———next-video is a library for adding video to Next.js applications. Prior to 2.8.1, the GET endpoint exported by next-video/request-handler and commonly mounted at /api/video accepts an unauthenticated url query parameter, while src/utils/utils.ts isRemote() treats any value without an HTTP or HTTPS prefix as a local path. src/request-handler.ts passes that value through src/assets.ts getAssetPath() to src/config.ts loadAsset(), which appends a JSON suffix and uses fs.readFile without canonicalizing the path or verifying that it remains inside the configured video folder. A remote attacker can therefore escape the intended asset directory and read JSON files accessible to the application process, including Next.js server-action encryption material, preview-mode keys, build manifests, route metadata, absolute paths, and application video asset identifiers. Applications that do not expose the runtime request handler are not affected through this route. This issue is fixed in version 2.8.1.5h CVE-2026-472565.3 MED

—

———OpenTelemetry, also known as OTel, is a vendor-neutral open source Observability framework for instrumenting, generating, collecting, and exporting telemetry data such as traces, metrics, and logs. Prior to 0.154.0, the Sentry exporter reads the remote OTLP sender-controlled service.name resource attribute in exporter/sentryexporter/sentry_exporter.go through extractProjectSlug and getOrCreateProjectEndpoint, passes the raw project slug to GetOTLPEndpoints and GetProjectKeys in exporter/sentryexporter/sentry_client.go, and interpolates it into a Sentry API URL without applying projectSlugRegexp through validateRoutingConfig at runtime in exporter/sentryexporter/config.go. Special characters can turn the expected path suffix into query data in all deployments or introduce slash and dot segments that traverse paths when the Sentry deployment normalizes them, while the Collector attaches its operator-configured bearer token to the request. A successful request can reach token-authorized administrative, organization, member, or key endpoints within the configured Sentry organization, and an attacker-controlled project slug can redirect subsequently exported telemetry. Sentry token middleware prevents cross-organization access. This issue is fixed in version 0.154.0.5h CVE-2026-571459.1 CRÍ

—

———PraisonAI is a multi-agent teams system. Prior to 4.6.62, src/praisonai/praisonai/tools/multiedit.py passes the LLM-controlled filepath parameter directly to open for reading and writing without traversal rejection, symlink resolution, a workspace boundary, or protected-path checks. Prompt-influenced agents can read files through edit and diff behavior or overwrite files accessible to the process, exposing secrets and enabling persistence or application tampering. This issue is fixed in 4.6.62.7h CVE-2026-571197.5 ALT

—

———PraisonAI is a multi-agent teams system. Prior to 4.6.59, the unauthenticated Jobs API accepts an absolute or traversing agent_file path in POST /api/v1/runs and passes it to the job executor without a workspace allowlist or boundary check. A remote caller can cause the server to open files accessible to the service account, exposing credentials, keys, environment variables, and other local data. This vulnerability is fixed in 4.6.59.7h