CVE Feed

The mamba language model framework thru 2.2.6 is vulnerable to insecure deserialization (CWE-502) when loading pre-trained models from HuggingFace Hub. The MambaLMHeadModel.from_pretrained() method uses torch.load() to load the pytorch_model.bin weight file without enabling the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by publishing a malicious model repository on HuggingFace Hub. When a victim loads a model from this repository, arbitrary code is executed on the victim's system in the context of the mamba process.

The imgaug library thru 0.4.0 contains an insecure deserialization vulnerability in its BackgroundAugmenter class within the multicore.py module. The class uses Python's pickle module to deserialize data received via a multiprocessing queue in the _augment_images_worker() method without any safety checks. An attacker who can influence the data placed into this queue (e.g., through social engineering, malicious input scripts, or a compromised shared queue) can provide a malicious pickle payload. When deserialized, this payload can execute arbitrary code in the context of the worker process, leading to remote or local code execution depending on the deployment scenario.

The mem0 1.0.0 server lacks authentication and authorization controls for its memory management API endpoints. Critical functions such as updating memory records (PUT /memories/{memory_id}) are exposed without any verification of the requester's identity or permissions. A remote attacker can exploit this by sending unauthenticated requests to modify, overwrite, or delete arbitrary memory records, leading to unauthorized data manipulation and potential data loss.

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization (CWE-502) through its predict() method. When a user provides a dataset file path to the predict() method, the framework automatically determines the file format. If the file is a pickle (.pkl) file, it is loaded using pandas.read_pickle() without any validation or security restrictions. This allows the deserialization of arbitrary Python objects via the unsafe pickle module. A remote attacker can exploit this by providing a maliciously crafted pickle file, leading to arbitrary code execution on the system running the Ludwig prediction.

The mem0 1.0.0 server lacks authentication and authorization controls for its memory creation API endpoint (POST /memories). The endpoint allows unauthenticated users to submit arbitrary memory records without verifying their identity or permissions. A remote attacker can exploit this by sending unauthenticated POST requests to create malicious or spoofed memory entries in the database, leading to unauthorized data injection and potential data pollution.

Guardrails AI thru 0.6.7 contains a code injection vulnerability (CWE-94) in its Hub package installation mechanism. When installing validator packages via guardrails hub install, the system retrieves a manifest from the Guardrails Hub and dynamically executes a script specified in the post_install field. The script path is constructed from untrusted manifest data and executed without proper validation or sanitization, allowing remote code execution. An attacker who can publish malicious packages to the Hub can inject arbitrary code that will be executed on any system where a victim installs the malicious package.

The CosyVoice project thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its model loading process. When loading model files (.pt) from a user-specified directory (via the --model_dir argument), the code uses torch.load() without the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the Pickle module. An attacker can exploit this by providing a maliciously crafted model directory containing .pt files with embedded pickle payloads. When a victim loads this directory using CosyVoice's web interface, the malicious payload is executed, leading to remote code execution on the victim's system.

The mem0 v1.0.0 server lacks authentication and authorization controls for its memory reset functionality accessible via the DELETE /memories endpoint. An unauthenticated attacker can send a DELETE request that triggers a reset operation, leading to the execution of a DROP TABLE SQL statement. This results in the deletion of the entire memory database table, causing catastrophic data loss and a complete denial of service for all users of the service.

The Adversarial Robustness Toolbox (ART) thru 1.20.1 contains a remote code execution vulnerability in its Kubeflow component. The robustness evaluation function for PyTorch models uses the unsafe eval() function to dynamically evaluate user-supplied strings for the LossFn and Optimizer parameters without any sanitization or security restrictions. An attacker can exploit this by providing a specially crafted string that contains arbitrary Python code, which will be executed when eval() is called, leading to complete compromise of the system running the ART evaluation.

The superduper project thru v0.10.0 contains a critical remote code execution vulnerability in its query parsing component. The _parse_op_part() function in query.py uses the unsafe eval() function to dynamically evaluate user-supplied query operands without proper sanitization or restriction. Although the function attempts to limit the execution context by providing a restricted global namespace, it does not block access to dangerous built-in functions. A remote attacker can exploit this by submitting a specially crafted query string containing Python code that imports modules (e.g., os) and executes arbitrary system commands, leading to complete compromise of the server.

An issue in Open Source Kubectl MCP Server v1.1.1 allows attackers to execute arbitrary code on a victim system via user interaction with a crafted HTML page.

The mem0 1.0.0 server lacks authentication and authorization controls for its memory deletion API endpoint (DELETE /memories). The endpoint allows unauthenticated users to delete memory records by specifying arbitrary user identifiers (e.g., user_id, run_id, agent_id) in the request query parameters. A remote attacker can exploit this by sending unauthenticated DELETE requests to erase memory data for any user, leading to unauthorized data loss and denial of service.

The Adversarial Robustness Toolbox (ART) thru 1.20.1 contains an insecure deserialization vulnerability (CWE-502) in its Kubeflow component's model loading functionality. When loading model weights from a file (e.g., model.pt) during robustness evaluation, the code uses torch.load() without the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the Pickle module. An attacker can exploit this by uploading a maliciously crafted model file to an object storage location referenced by the pipeline, or by controlling the model_id parameter to point to such a file. When the pipeline loads the model, the malicious payload is executed, leading to remote code execution.

The snorkel library thru v0.10.0 contains a critical insecure deserialization vulnerability (CWE-502) in the BaseLabeler.load() method of the BaseLabeler class. The method loads serialized labeler models using the unsafe pickle.load() function on user-supplied file paths without any validation or security controls. Python's pickle module is inherently dangerous for deserializing untrusted data, as it can execute arbitrary code during the deserialization process. A remote attacker can exploit this by providing a maliciously crafted pickle file, leading to arbitrary code execution on the victim's system when the file is loaded via the vulnerable method.

The torch-checkpoint-shrink.py script in the ml-engineering project in commit 0099885db36a8f06556efe1faf552518852cb1e0 (2025-20-27) contains an insecure deserialization vulnerability (CWE-502). The script uses torch.load() to process PyTorch checkpoint files (.pt) without enabling the security-restrictive weights_only=True parameter. This oversight allows the deserialization of arbitrary Python objects via the pickle module. A remote attacker can exploit this by providing a maliciously crafted checkpoint file, leading to arbitrary code execution in the context of the user running the script.

An insecure direct object reference in MK-Auth 23.01K4.9 allows attackers to access and send support calls for other users via manipulation of the chamado parameter through a crafted GET request.

The mem0 1.0.0 server lacks authentication and authorization controls for its memory reset and table re-creation functionality accessible via the DELETE /memories endpoint. An unauthenticated attacker can send a DELETE request that triggers a reset operation, leading to the execution of a CREATE TABLE SQL statement. This can cause unexpected table re-creation, schema disruption, potential data loss, and denial of service for the memory management service.

The TinyZero project thru commit 6652a63c57fa7e5ccde3fc9c598c7176ff15b839 (2025-58-24) contains a critical command injection vulnerability (CWE-78) in its HDFS file operation utilities. The vulnerability arises from the unsafe construction and execution of shell commands via os.system() without proper input sanitization or escaping. User-controlled input (such as file paths) is directly interpolated into shell command strings using f-strings within the _copy() function. An attacker can inject arbitrary OS commands by supplying a specially crafted path parameter through the Hydra configuration framework. This leads to remote code execution with the privileges of the user running the TinyZero training process.

A Stored Cross-Site Scripting (XSS) vulnerability was discovered in the File Management module of FluentCMS 1.2.3. The flaw allows an authenticated administrator to upload crafted SVG files containing malicious JavaScript code. Once uploaded, the script executes in the browser of any user who accesses the direct URL of the image, including unauthenticated visitors.

Exim before 4.99.3, in certain GnuTLS configurations, has a remotely reachable use-after-free in the BDAT body parsing path. It is triggered when a client sends a TLS close_notify mid-body during a CHUNKING transfer, followed by a final cleartext byte on the same TCP connection. This can lead to heap corruption. An unauthenticated network attacker exploiting this vulnerability could execute arbitrary code.

The mem0 1.0.0 server lacks authentication and authorization controls for its memory deletion API endpoint (DELETE /memories/{memory_id}). The endpoint allows unauthenticated users to delete arbitrary memory records without verifying their identity or permissions. A remote attacker can exploit this by sending unauthenticated DELETE requests to remove any memory entry from the database, leading to unauthorized data loss and potential denial of service.

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization (CWE-502) in its model serving component. When starting a model server with the ludwig serve command, the framework loads model weight files using torch.load() without enabling the security-restrictive weights_only=True parameter. This default behavior allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing a maliciously crafted PyTorch model file, leading to arbitrary code execution on the system hosting the Ludwig model server.

The llm CLI tool thru 0.27.1 contains a critical code injection vulnerability via its --functions command-line argument. This argument is intended to allow users to provide custom Python function definitions. However, the tool directly executes the provided code using the unsafe exec() function without any sanitization, sandboxing, or security restrictions. An attacker can exploit this by crafting a malicious llm command with arbitrary Python code in the --functions argument and using social engineering to trick a victim into running it. This leads to arbitrary code execution on the victim's system, potentially granting the attacker full control.

Horovod thru 0.28.1 contains an insecure deserialization vulnerability (CWE-502) in its KVStore HTTP server component. The KVStore server, used for distributed task coordination, lacks authentication and authorization controls, allowing any remote attacker to write arbitrary data via HTTP PUT requests. When a Horovod worker reads data from the KVStore (via HTTP GET), it deserializes the data using cloudpickle.loads() without verifying its source or integrity. An attacker can exploit this by sending a malicious pickle payload to the server before the legitimate data is written, causing the victim worker to deserialize and execute arbitrary code, leading to remote code execution.

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability (CWE-502) in the MultitaskClassifier.load() method of the MultitaskClassifier class. The method loads model weight files using torch.load() without enabling the security-restrictive weights_only=True parameter. This default behavior allows the deserialization of arbitrary Python objects via the Pickle module. A remote attacker can exploit this by providing a maliciously crafted model file, leading to arbitrary code execution on the victim's system when the file is loaded via the vulnerable method.

PySyft (Syft Datasite/Server) versions 0.9.5 and earlier are vulnerable to remote code execution due to insufficient validation and sandboxing of user-submitted code. The system allows low-privileged users to submit Python functions (via @sy.syft_function()) for remote execution on the server. While a code approval mechanism exists, the submitted code undergoes no security checks for dangerous operations (e.g., file access, command execution). Once approved, the code is executed within the server process using exec() and eval() functions without proper isolation. A remote attacker can leverage this to execute arbitrary Python code on the server, leading to complete compromise of the server environment.

The _load_model() function in the neural_magic_training.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f (2024-07-21) is vulnerable to insecure deserialization (CWE-502). When loading a model state dictionary from a state_dict.pt file via torch.load(), the function does not enable the weights_only=True security parameter. This allows the deserialization of arbitrary Python objects through the Pickle module. A remote attacker can exploit this by providing a maliciously crafted state_dict.pt file within a directory specified via the --model argument, leading to arbitrary code execution during the deserialization process on the victim's system.

The nexent v1.7.5.2 backend service contains an unauthorized arbitrary storage file deletion vulnerability in its file management API. The DELETE /storage/{object_name:path} endpoint lacks authentication, authorization, and input validation mechanisms. Unauthenticated remote attackers can send crafted requests with a user-controlled object_name path parameter to delete arbitrary files from the underlying MinIO storage system. Successful exploitation leads to data loss and denial of service.

The nexent v1.7.5.2 backend service contains an unauthorized arbitrary file deletion vulnerability in its ElasticSearch service interface. The DELETE /{index_name}/documents endpoint lacks proper authentication and authorization controls and does not validate the user-supplied path_or_url parameter. This allows unauthenticated remote attackers to send crafted requests that trigger the deletion of arbitrary documents from ElasticSearch indices and corresponding files from the MinIO storage system. Successful exploitation leads to data destruction and denial of service.

A flaw has been found in omec-project amf up to 2.1.1. This vulnerability affects unknown code of the component NGAP Message Handler. Executing a manipulation can lead to memory corruption. The attack can be launched remotely. The exploit has been published and may be used. This patch is called 8a4c33cdda866094f1989bdeff6d8642fce8de8435f89defd66831c97715f5aa. It is best practice to apply a patch to resolve this issue.

A vulnerability was detected in D-Link DIR-816 1.10CNB05_R1B011D88210. This affects the function portForward. Performing a manipulation of the argument ip_address results in command injection. The attack can be initiated remotely. The exploit is now public and may be used.

barebox version prior to 2026.04.0 contains multiple memory-safety vulnerabilities in the EFI PE loader in efi/loader/pe.c where integer overflow in virtual image size computation using 32-bit arithmetic on section VirtualAddress and size values allows undersized heap allocation, and PE section loading logic fails to validate that PointerToRawData plus copied size remains within the PE file buffer. An attacker can supply a malicious EFI PE binary via TFTP, USB, SD card, or network boot to trigger heap buffer overflow or out-of-bounds read from heap memory, potentially achieving code execution in bootloader context.

Vaultwarden is a Bitwarden-compatible server written in Rust. Prior to 1.35.4, there is a security vulnerability in Vaultwarden that allows bypassing the login brute-force protection if email 2fa is enabled. If email 2fa is enabled, the unprotected 2fa-function send_email_login (email.rs, api endpoint /api/two-factor/send-email-login) also acts as an oracle determining whether a username-password combination is correct. An attacker can abuse that endpoint to brute-force passwords without rate-limiting. This works even for users who don't have email 2fa configured. This vulnerability is fixed in 1.35.4.

Vaultwarden is a Bitwarden-compatible server written in Rust. Prior to 1.35.5, Vaultwarden allows an unconfirmed organization owner to purge the entire organization vault. The organization invite flow uses a two-step process: accepting an invite transitions membership from Invited to Accepted, and a separate confirmation by an existing owner upgrades it to Confirmed. The POST /api/ciphers/purge endpoint uses plain Headers and only checks that the membership type is Owner without verifying that the membership status is Confirmed. An authenticated user who has been invited as an organization owner and has accepted the invite and has not yet been confirmed can call this endpoint to hard-delete all ciphers and attachments in the organization, causing immediate organization-wide data loss. This vulnerability is fixed in 1.35.5.

A security vulnerability has been detected in D-Link DIR-816 1.10CNB05_R1B011D88210. Affected by this issue is the function sub_445E7C of the file /goform/singlePortForward. Such manipulation of the argument ip_address leads to command injection. It is possible to launch the attack remotely. The exploit has been disclosed publicly and may be used.

barebox version prior to 2026.04.0 contains a denial-of-service vulnerability in ext4 directory parsing in fs/ext4/ext4_common.c where the ext4fs_iterate_dir() function fails to validate that directory entry length values are non-zero. Attackers can supply a malicious ext4 filesystem image with a crafted directory entry containing a direntlen value of 0 to cause an infinite loop during directory listing or path resolution, resulting in the boot process hanging indefinitely.

Vaultwarden is a Bitwarden-compatible server written in Rust. Prior to 1.35.5, Vaultwarden does not enforce that a groups_users.users_organizations_uuid entry belongs to the same organization as groups.groups_uuid, or a collections_groups.collections_uuid entry belongs to the same organization as collections_groups.groups_uuid. Multiple organization group-management endpoints accept arbitrary MembershipId and CollectionId values and persist them directly without verifying org consistency. This lets an attacker who is Admin in Organization A, and only a low-privileged member in Organization B bind their Org B membership UUID into an Org A group, then use that foreign group relationship to gain unauthorized access to Org B vault data. With an accessAll=true Org A group, the attacker can make /api/sync and /api/ciphers enumerate Org B ciphers. Once those unauthorized sync results reveal Org B collection IDs, the attacker can also bind those foreign collection IDs to the Org A group and turn the same flaw into write access over Org B items. This vulnerability is fixed in 1.35.5.

Vaultwarden is a Bitwarden-compatible server written in Rust. Prior to 1.35.5, refresh tokens are not invalidated when the user's security_stamp is rotated by some security-sensitive operations (password change, KDF change, key rotation, email change, org admin password reset, emergency access takeover). This allows an attacker holding a previously obtained refresh token to maintain session access even after the user has taken action to secure their account. This vulnerability is fixed in 1.35.5.

Wireshark MCP is an MCP Server that turns tshark into a structured analysis interface, then layers in optional Wireshark suite utilities. In 1.1.5 and earlier, wireshark-mcp exposes a wireshark_export_objects MCP tool that accepts an attacker-controlled dest_dir parameter and passes it to tshark's --export-objects flag with no mandatory path restriction. The path sandbox (_allowed_dirs) is None by default and only activates when the environment variable WIRESHARK_MCP_ALLOWED_DIRS is explicitly set. In a default installation, any directory on the filesystem can be used as the export destination.

Fiber is a web framework for Go. Prior to 2.52.12 and 3.1.0, Cross-Site Scripting vulnerability in Go Fiber allows a remote attacker to inject arbitrary HTML/JavaScript by supplying Accept: text/html on any request whose handler passes attacker-influenced data to the AutoFormat() feature. The developer opts into content negotiation by calling AutoFormat(), but does not opt into raw HTML emission for a particular request; Fiber chooses that branch from attacker-controlled Accept. The html branch is the sole outlier in a method whose name (AutoFormat) and symmetrical structure actively telegraph "safe, format-agnostic reply." This vulnerability is fixed in 2.52.12 and 3.1.0.

DeepChat is an open-source artificial intelligence agent platform that unifies models, tools, and agents. Prior to v1.0.4-beta.1, An incomplete mitigation for CVE-2025-55733 leaves DeepChat vulnerable to an arbitrary protocol execution bypass (RCE). While the patch correctly restricted api.openExternal() inside the renderer's preload/index.ts script, it structurally neglected to sanitize native Electron pop-up window handlers. An attacker or a compromised AI endpoint returning a Markdown link can trigger a target="_blank" native window interception in tabPresenter.ts, which forwards the malicious URL directly to shell.openExternal(url) and completely bypasses the isValidExternalUrl security boundary. This vulnerability is fixed in v1.0.4-beta.1.

DeepChat is an open-source artificial intelligence agent platform that unifies models, tools, and agents. Prior to v1.0.4-beta.1, a Cross-Site Scripting (XSS) vulnerability exists due to a discrepancy between the backend validation layer and the frontend browser rendering engine. The SVGSanitizer (src/main/lib/svgSanitizer.ts) restricts script execution by scrubbing javascript: protocols using plain-text regular expressions. However, it fails to account for HTML entity decoding prior to Vue's v-html DOM insertion inside the SvgArtifact.vue component. By feeding an SVG artifact with obfuscated entities (e.g., javascript:alert(1)), an attacker can completely bypass the sanitizer, culminating in arbitrary JavaScript execution when a victim interacts with the rendered SVG Element. This vulnerability is fixed in v1.0.4-beta.1.

A weakness has been identified in D-Link DIR-816 1.10CNB05_R1B011D88210. Affected by this vulnerability is the function sub_445E7C of the file /goform/formDMZ.cgi. This manipulation causes command injection. It is possible to initiate the attack remotely. The exploit has been made available to the public and could be used for attacks.

Geyser is a bridge between Minecraft: Bedrock Edition and Minecraft: Java Edition. Prior to 2.9.3, a server-side request forgery (SSRF) vulnerability exists in Geyser’s handling of Bedrock player head texture data. By supplying a crafted Base64-encoded skin texture URL via the /give command, an attacker can cause the Minecraft server to issue arbitrary HTTP GET requests to attacker-controlled or internal endpoints. This occurs server-side, without proper URL validation, and can be triggered by a Bedrock client. This vulnerability is fixed in 2.9.3.

libcaca is a colour ASCII art library. In 0.99.beta20 and earlier, an integer overflow vulnerability in libcaca's canvas import functionality allows an attacker to cause a controlled heap out-of-bounds write (heap overflow) by supplying a crafted file in the "caca" format. Depending on the build configuration and memory allocator, this may lead to memory corruption or remote code execution. This is the same vulnerability as CVE-2021-3410 but the fix at that time was not fully correct. Commit fb77acff9ba6bb01d53940da34fb10f20b156a23 fixes this vulnerability.

jotty·page is a self-hosted app for your checklists and notes. Prior to 1.22.0, an unauthenticated path traversal vulnerability exists in /api/app-icons/[filename]. The filename route parameter is joined into a filesystem path without traversal/boundary validation, allowing file reads outside data/uploads/app-icons/. This vulnerability is fixed in 1.22.0.

Link Preview JS extracts web links information. Prior to 4.0.1, the library did not check for IPv6 loopback attacks. There was also a DNS attack, where an address could be resolved into an internal IP. This could cause internal data leaks. This vulnerability is fixed in 4.0.1.

HTTP::Tiny versions before 0.093 for Perl do not validate CRLF in HTTP request lines or control field header values. The unvalidated inputs are the method and URI in the request line, the URL host that becomes the `Host:` header, and HTTP/1.1 control data field values. An attacker who controls one of these inputs, for example a user supplied URL passed to a webhook or URL fetch endpoint, can inject additional headers and smuggle requests to the upstream server.

Outline is a service that allows for collaborative documentation. Prior to 1.7.1, the Slack integration callback for GET /auth/slack.post accepts an unsigned, session-independent OAuth state value. A third party who can obtain a Slack OAuth code for the same Outline Slack client can make a logged-in Outline user complete the callback and link that user's Outline account to the attacker's Slack team_id and user_id. The linked Slack identity can then use the Slack /outline search command as the victim Outline user. This vulnerability is fixed in 1.7.1.

Outline is a service that allows for collaborative documentation. Prior to 1.7.0, the shares.create API accepts both collectionId and documentId simultaneously and, when published=false, only verifies read access for each—skipping the "share" permission check. A subsequent shares.update authorizes publication using an OR policy (can share collection OR can share document), so an attacker who holds share permission on one unrelated collection can publish a share that exposes an arbitrary document they cannot legitimately share, making it publicly accessible to unauthenticated users. This vulnerability is fixed in 1.7.0.