| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| QWED-MCP is a deterministic verification gateway for MCP. Prior to 0.2.1, verify_math_expression() in src/qwed_mcp/engines/math_engine.py passes attacker-controlled expression and claimed_result strings directly to SymPy's parse_expr() after only normalizing caret syntax to Python exponent syntax, without restricting global_dict, removing Python built-ins, or validating the expression AST. Because parse_expr() calls Python's eval() with built-ins available, an attacker who can cause a downstream caller to pass untrusted input to this public library function can use Python import functionality to execute arbitrary operating-system commands as the qwed-mcp process user, read or modify accessible data, exfiltrate process secrets, or reach internal services. The default MCP tool registry does not expose verify_math_expression(), so exploitation requires a downstream integration that invokes the library API with attacker-controlled input. This issue is fixed in version 0.2.1. |
| Vulnerability in Drupal Powerful Surveys. This issue affects Powerful Surveys versions: *.*. |
| CorvusSKK contains a code injection vulnerability, which may lead to arbitrary code execution on the affected product. |
| whichllm before 0.5.16 contains a code injection vulnerability in the run and snippet commands that allows a remote attacker who controls a HuggingFace repository to achieve arbitrary code execution by crafting a malicious GGUF filename containing double quotes or other special characters. The script generation function in cli.py interpolates HuggingFace-derived values, including GGUF variant filenames from the Hub API siblings rfilename field, directly into Python source code without escaping, allowing the crafted filename to break out of the generated string literal and execute injected code on the user's machine before any model download occurs. |
| Trilium is an open-source hierarchical note-taking application. In versions up to and including 0.103.0, the default-on "Safe import" filter does not sanitize note titles, and the GeoMap note view interpolates a marker note's title into raw HTML that is rendered as innerHTML, allowing an attacker-supplied import archive to inject script that runs when the map is displayed. Because Safe import neutralizes scripts but never escapes titles, a note whose title contains an HTML event-handler payload survives the import and executes as soon as the victim opens the GeoMap that renders its marker. On the desktop client the Electron renderer runs with Node integration enabled, so the injected JavaScript escalates from cross-site scripting to full remote code execution on the victim's machine. This issue is fixed in version 0.104.0. |
| Cohere North AI v1.1.5 was discovered to contain excessively permissive cross-domain policy with untrusted domains. This occurs via the server failing to validate the Origin header of incoming connection requests. |
| Diffusers is the a library for pretrained diffusion models. Prior to 0.38.0, a trust_remote_code bypass in DiffusionPipeline.from_pretrained allows arbitrary remote code execution despite the user passing trust_remote_code=False (or omitting it, which is the default). The vulnerability has three variants, all sharing the same root cause — the trust_remote_code gate was implemented inside DiffusionPipeline.download() rather than at the actual dynamic-module load site, so any code path that bypassed or short-circuited download() also bypassed the security check. DiffusionPipeline.from_pretrained('repoA', custom_pipeline='attacker/repoB', trust_remote_code=False) — the gate evaluated against repoA's file list rather than repoB's, so repoB's pipeline.py was loaded and executed. DiffusionPipeline.from_pretrained('/local/snapshot', custom_pipeline='attacker/repoB', trust_remote_code=False) — the local-path branch never invoked download(), so the gate was never reached and remote code from repoB executed. DiffusionPipeline.from_pretrained('/local/snapshot', trust_remote_code=False) where the snapshot contains custom component files (e.g. unet/my_unet_model.py) referenced from model_index.json — same root cause; the local path skipped download() and custom component code executed. This vulnerability is fixed in 0.38.0. |
| protobufjs compiles protobuf definitions into JavaScript (JS) functions. Prior to 7.5.6 and 8.0.2, protobufjs generated JavaScript for toObject conversion could include an unsafe expression derived from a schema-controlled bytes field default value. A crafted descriptor with a non-string default value for a bytes field could cause attacker-controlled code to be emitted into the generated conversion function. This vulnerability is fixed in 7.5.6 and 8.0.2. |
| ToolUniverse ran caller-supplied Python inside a sandbox that could be escaped, on a server that required no authentication. The executor behind the python_code_executor tool, in python_executor_tool.py, inspected the submitted source for a denied list of attribute names and calls but left the attribute-lookup builtins available and did not stop a dunder attribute reached through a string lookup or through a module already permitted, so a caller could walk from a literal's class to its base and enumerate subclasses to obtain a reference to the process and subprocess modules. A per-call argument also let the caller widen the import allow-list before the inspection ran. The HTTP and MCP servers in http_api_server.py and smcp_server.py bound to every interface with debugging enabled and no authentication, so any caller able to reach the port executed code as the server process. Version 1.3.0 adds bearer-token authentication, defaults the bind address to loopback, and hardens the attribute checks. |
| OpenMed before 1.5.2 contains a remote code execution vulnerability in the PII privacy-filter model loading path. The privacy-filter dispatcher used broad substring matching on the user-supplied model_name parameter, allowing a value such as attacker/foo-privacy-filter-bar to route through a path that loads Hugging Face models with trust_remote_code=True. An unauthenticated attacker can supply a malicious model repository containing custom Transformers code via auto_map in config.json or tokenizer_config.json, which is imported and executed with the privileges of the OpenMed service process. |
| Python StateMachine versions 3.0.0 before 3.2.0 contains a remote code execution vulnerability that allows attackers to execute arbitrary code by supplying malicious SCXML documents containing crafted `<data expr="...">` attributes evaluated unsafely. The SCXMLProcessor passes attacker-controlled expression strings through a call chain ending in Python's built-in eval() without sandboxing, enabling arbitrary code execution in the context of the hosting process. |
| uproot dynamically generates Python class source code from ROOT TStreamerInfo records in a file and compiles it at runtime. Some file-controlled streamer metadata fields (for example, streamer element names) are interpolated into the generated Python source without safe quoting via repr() or the !r format specifier. An attacker who can supply a crafted ROOT file can place Python expression-breaking content into a streamer metadata field. When uproot generates and invokes the corresponding reader method, the injected Python expression is evaluated in the context of the process opening the file, resulting in arbitrary Python code execution in applications that open or process attacker-controlled ROOT files with affected uproot code paths. |
| Trilium is an open-source hierarchical note-taking application. In versions up to and including 0.103.0, the default-on "Safe import" filter sanitizes HTML only for text notes and excludes the book note type, whose content is stored without sanitization and later rendered as HTML, allowing an attacker-supplied import archive to embed a payload that executes as script. A book note's content is routed through the same rendering path as text notes and injected into the page with jQuery's html method when the note is shown as a grid-view preview card, so a malicious note survives Safe import and runs as soon as the victim opens the containing note. On the desktop client the Electron renderer runs with Node integration enabled, so the injected JavaScript escalates from cross-site scripting to full remote code execution on the victim's machine. This issue is fixed in version 0.104.0. |
| Budibase versions before 3.41.3 contain a remote code execution vulnerability in plugin handling that allows authenticated admin users to execute arbitrary code by uploading a malicious plugin tarball. The server calls eval() on plugin JavaScript files without sandboxing in the main Node.js process, enabling attackers to exfiltrate environment variables and credentials with root privileges in default deployments. |
| Code injection in Bisection in Google Chrome prior to 152.0.7977.65 allowed a remote attacker to obtain sensitive information via a crafted file. (Chromium security severity: Medium) |
| A security vulnerability has been detected in RooCodeInc Roo-Code up to 3.51.1. This affects the function fetch_instructions of the file malicious_mcp_server.py of the component MCP Integration Trust Model. The manipulation leads to code injection. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used. Multiple isses were reported to the vendor beforehand. They explain, that "they all apply to Roo Code, a project we no longer support - the repository was archived a while ago, and we don't encourage anyone to use it." This vulnerability only affects products that are no longer supported by the maintainer. |
| A weakness has been identified in primefaces primereact up to 10.9.8. This issue affects the function ObjectUtils.mutateFieldData of the file components/lib/utils/ObjectUtils.js of the component API. This manipulation of the argument Field causes improperly controlled modification of object prototype attributes. The attack is possible to be carried out remotely. Upgrading to version 10.9.9 and 11.0.0 is capable of addressing this issue. Patch name: 61f182e11d9ef52032ff56f420da763a6938236f. It is recommended to upgrade the affected component. |
| SQL injection in Hive Metastore direct SQL partition-name resolution in Apache Hive before 4.2.1 on all platforms allows authenticated users with access to Hive Metastore APIs to read, modify, or affect unintended partition metadata (including statistics updates, truncation targets, and file-metadata cache operations) via crafted partition names in metastore RPC requests when direct SQL is enabled (the default). Users are recommended to upgrade to version 4.2.1, which fixes this issue.
Details about the issue:
Several Hive Metastore RPCs resolve partitions by full partition name (PART_NAME) through direct-SQL helpers. In those paths, client-supplied partition names are embedded into SQL using string concatenation (DirectSqlUpdatePart.quoteString() → '...') instead of bind parameters. A partition name containing a single quote (and crafted SQL) can alter the generated WHERE clause so that lookups intended for one partition match additional rows. That can affect reads, stats updates, truncate targets, metadata-cache targets, and related operations when metastore.try.direct.sql is enabled (default: true). An authenticated or network-trusted caller with the ability to invoke Hive Metastore partition-name APIs against a target table (directly or via Hive/other clients), when direct SQL is enabled can perform this attack. Also, the impact is mainly within table & partition targeting (read/update/truncate/drop/cache the wrong partitions in a table they can reference), not arbitrary cross-database access via this bug alone. |
| A vulnerability was determined in jQWidgets up to 24.0.1. This affects the function JQXLite.extend/jqxBaseFramework.extend of the file jqwidgets/jqx-all.js. This manipulation causes improperly controlled modification of object prototype attributes. The attack can be initiated remotely. The reported GitHub issue was closed with the label "not planned". |
| In Splunk Enterprise versions below 10.4.2, 10.2.6, 10.0.9, and 9.4.14, an authenticated user who does not hold a role with the edit_manager_xml capability could write a malicious Splunk Web Manager Extensible Markup Language (XML) configuration. When the same user opens the affected Splunk Web Manager page, Splunk Enterprise runs attacker-controlled operating-system commands as the user account running Splunk Enterprise. The vulnerability is possible because Splunk Web does not require the edit_manager_xml capability before accepting Splunk Web Manager XML configuration changes. |
