CVE-2026-116691
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 2.19.0 until 3.17.0, m
CVSS
6.5
Medium
EPSS
—
KEV
—
Exploit Today
0
0-100
Published: Oct 27, 2026 · Last modified: Oct 28, 2026 · CWE-200 · CWE-312 · CWE-532
Not enough EPSS history yet.
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 2.19.0 until 3.17.0, mlflow.crewai.autolog() records CrewAI Agent objects without consistently sanitizing nested credential-bearing values. mlflow/crewai/autolog.py::_get_agent_attributes() converts the Agent llm field to a string that includes the provider API key, while mlflow/crewai/autolog.py::_construct_full_inputs() expands the Agent object into mlflow.spanInputs. Applications can therefore send plaintext provider credentials to the MLflow trace backend or a configured OTLP destination, where trace readers can obtain them. This issue is fixed in version 3.17.0.
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-1185784.3 MED—
———A security flaw has been discovered in maotoumao MusicFreeDesktop 0.0.6/0.0.7/0.0.8. The affected element is an unknown function of the component Request Forwarder. The manipulation results in information disclosure. The attack can only be performed from the local network. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.4hCVE-2026-118666——
———GraphQL Tools provides utilities for building, stitching, and mocking GraphQL schemas. Prior to version 1.1.36 of the GraphQL Tools legacy WebSocket executor package, buildWSLegacyExecutor in packages/executors/legacy-ws/src/index.ts reuses one WebSocket and replaces its single websocket.onmessage handler whenever a new overlapping subscription starts. The handler delivers LEGACY_WS.DATA frames to the newest observer without comparing data.id with the subscription operation id, so a caller using the same executor can receive another user's authorized subscription payload while the earlier caller's stream is starved. This issue is fixed in version 1.1.36 of the GraphQL Tools legacy WebSocket executor package.6hCVE-2026-631909.4 CRI—
———FearlessCMS is a flat-file `content` management system centered on simplicity and open source. Prior to version 0.1.4, five debug/test PHP scripts under `admin/` perform privileged operations without `isLoggedIn()` authentication checks, CSRF protection or adequate `path` validation. On PHP's built-in development server, including the default Docker deployment, `router.php` checks an unnormalized admin URI prefix while the filesystem resolves dot segments, so `/./admin/` bypasses the admin login gate and executes those scripts directly. In `admin/debug_save_flow.php`, attacker-controlled `path` and `content` reach `file_put_contents` for arbitrary `.md` writes under `content`/ or outside it through `path` traversal, enabling site defacement and stored JavaScript in public pages; `print_r($_POST)` output without `htmlspecialchars` escaping also causes reflected cross-site scripting. The `admin/debug_content.php` `path` parameter reaches `scandir()`, `file_get_contents()` and `posix_getpwuid()`, permitting arbitrary directory enumeration and `.md` file reads that expose absolute paths, ownership, permissions, hashes and `content` previews. The `admin/preview-handler.php` endpoint accepts JSON `content` and `title` and writes enumerable files under `content/_preview/`, enabling stored cross-site scripting and disk exhaustion. The `admin/testdebug.php` endpoint appends to `admin/testdebug.log` without authentication, also enabling disk exhaustion. The `admin/test_redirect.php` script exposes a test artifact but escapes its inputs and is not itself an XSS sink. Stored script execution can steal sessions and compromise administrators when they view affected `content`. Reads and writes execute with PHP's privileges; the tested deployment runs PHP as root, amplifying filesystem impact. This issue is fixed in version 0.1.4.7hCVE-2026-1186478.2 HIG—
———n8n before 1.123.85, 2.x before 2.43.5, and 2.44.x before 2.44.1 saves the SeaTable Trigger node's API token in error execution data, because failed poll requests store raw request headers. Users who can view workflow executions but cannot read the credential can recover the token and access the connected SeaTable base.8hCVE-2026-796566.0 MED—
———A user with read access to Grafana's server logs, or to a log store they are shipped to, can recover the OAuth and OIDC access, refresh and ID tokens of users who sign in through the Generic OAuth, Google or GitLab providers. Grafana writes these tokens to its logs unredacted when an access or ID token is not a valid JWT, when the ID token is missing (at debug level) or is not a string (exposure of sensitive information in log files). The exposure is bounded by the scopes granted to the configured OAuth client.9hCVE-2026-1056176.5 MED—
———PostgreSQL Anonymizer doesn't invalidate plans prepared after a rule was added or a role changed. Users using such plans can still see the unmasked data until the end of the session. The problem is resolved in PostgreSQL Anonymizer 3.3.0 and later versions.9h