Veille CVE

Jeudi 23 juillet 2026

23 CVE

Critique 10 Élevée 9 Moyenne 4 Faible 0 Non évaluée 0 Exploitées (KEV) Tri : plus récentes « Python python » ✕
  1. CVE-2026-8838 9.8 Critique

    AWS Amazon Redshift connector for Python

    Unsafe use of Python's eval() on server-received data in the vector_in() function in amazon-redshift-python-driver before 2.1.14 allows a rogue server or man-in-the-middle actor to execute arbitrary code on the client. To remediate this issue, users should upgrade to version 2.1.14.

  2. CVE-2026-47708 9.3 Critique

    SepineTam stata-mcp

    MCP-for-Stata is an MCP server for Stata to integrate Stata into an agent. Prior to version 1.17.3, the `log_file_name` parameter in the `stata_do` API and CLI is directly interpolated into a Stata command string without sanitization. The security guard (`GuardValidator`) only scans the do-file content but does not validate this parameter. An attacker can inject arbitrary Stata commands (including `shell`, `python`, `erase`, etc.) by crafting a malicious `log_file_name` containing quotes, newlines, or Stata command separators. Version 1.17.3 contains a patch for the issue.

  3. CVE-2026-47731 9.1 Critique

    NASA-AMMOS AIT-Core

    The AMMOS Instrument Toolkit (Formerly the Bespoke Links to Instruments for Surface and Space (BLISS)) is a Python-based software suite developed to handle Ground Data System (GDS), Electronic Ground Support Equipment (EGSE), commanding, telemetry uplink/downlink, and sequencing for instrument and CubeSat Missions. In versions prior to 2.6.1 and in version 3.1.0, the Binary Stream Capture (BSC) component exposes an unauthenticated HTTP API for dynamically creating packet capture "handlers." Because the code blindly trusts path‑related form fields, a remote client can bypass the configured log root and direct BSC to log to arbitrary filesystem paths (path traversal / directory escape), and append attacker‑controlled data to those files, using the privileges of the`ait-bsc` process. There are two ways for a remote attacker to trigger this. First, if the attacker has access to the network where `ait-bsc` is deployed (a reason for that could be that the ports are publicly accessible), the payloads can be directly sent to the server to trigger the arbitrary file append. This type of attack is demonstrated in `python_poc.py`. Second, even if the attacker does not have direct access to the network because the software is running in a local network, it is possible to exploit this if a bad actor in that network opens an attacker-controlled website (which might be a website created by an attacker, or a third-party website compromised by the attacker). The browser javascript can automatically send the requests necessary to exploit this into the local network. This is even possible if the server is only accessible on `localhost`. This type of attack is demonstrated by `attacker_tcp.py` and `test1.html` (first launch the attacker TCP server, then start a webserver to host `test1.html`, for example using `python3 -m http.server 7000`, and open `test1.html`).This issue affects BSC (Binary Stream Capture) and usage of the ait-bsc server. This impacts AIT-Core versions before 3.1.1, from 2.x before 2.6.1. Users are recommended to upgrade to version 3.1.1 or 2.6.1.

  4. CVE-2026-50271 7.5 Élevée

    DataDog dd-trace-py

    Datadog dd-trace-py is the Datadog Python APM client. Prior to 4.8.2, Datadog tracing libraries that implement W3C baggage propagation parse incoming baggage HTTP headers without enforcing DD_TRACE_BAGGAGE_MAX_ITEMS or DD_TRACE_BAGGAGE_MAX_BYTES limits on the extract path. A remote, unauthenticated attacker can send a request whose baggage header contains an arbitrarily large number of comma-separated key-value pairs or a single very large value, causing unbounded CPU and memory consumption and enabling a remote denial of service against HTTP services with baggage propagation enabled. This issue is fixed in version 4.8.2.

  5. CVE-2026-47425 6.9 Moyenne

    conda rattler · conda py-rattler · conda rattler-build · prefix-dev pixi

    Rattler is a library that provides common functionality used within the conda ecosystem. Prior to version 0.43.2, `EntryPoint::FromStr` in `rattler_conda_types` performs only `.trim()` on the `command` field before the linker joins it onto the install prefix and writes an executable Python script. A malicious `noarch:python` package can ship an `info/link.json` with an entry-point name containing `..`, `/`, `\`, or an absolute path; the resulting file is written outside the prefix (or clobbers an existing in-prefix entry-point such as `bin/pip`) with mode `0o775` on Unix and a copied launcher `.exe` on Windows. This affects the default install path of `pixi install`, `mamba install` via py-rattler, `rattler-build`, and any other consumer of the `rattler` install crate; no flag or post-link-script opt-in is involved. Version 0.43.2 contains a fix for the issue.

  6. CVE-2026-49852 8.7 Élevée

    authlib joserfc

    joserfc is a Python library that provides an implementation of several JSON Object Signing and Encryption (JOSE) standards. Prior to 1.6.8, joserfc.jwt.decode accepts attacker-forged HMAC-signed tokens when the caller-supplied verification key is the empty string or None, because HMACAlgorithm.sign and HMACAlgorithm.verify in src/joserfc/_rfc7518/jws_algs.py pass the output of OctKey.get_op_key(...) to hmac.new(...) and OctKey.import_key in src/joserfc/_rfc7518/oct_key.py only emits a SecurityWarning for keys shorter than 14 bytes without rejecting zero-length input. This issue is fixed in version 1.6.8.

  7. CVE-2026-4372 7.8 Élevée

    huggingface/transformers · Huggingface transformers

    A critical remote code execution vulnerability exists in all versions of the HuggingFace transformers library prior to version 5.3.0. The vulnerability allows an attacker to craft a malicious `config.json` file containing the `_attn_implementation_internal` field set to an attacker-controlled HuggingFace Hub repository ID. When a victim loads this model using the standard `AutoModelForCausalLM.from_pretrained()` API, the library downloads and executes arbitrary Python code from the attacker's repository with the victim's full OS privileges. This issue arises due to unfiltered deserialization of configuration attributes, insufficient sanitization of internal fields, and unsandboxed execution of downloaded kernels. The vulnerability bypasses the `trust_remote_code` security mechanism, is invisible to the victim, and exploits the standard documented usage pattern, making it particularly severe. Users are advised to upgrade to version 5.3.0 or later to mitigate this issue.

  8. CVE-2026-48207 9.8 Critique

    Apache Software Foundation Apache Fory · Apache fory

    Deserialization of untrusted data in Apache Fory PyFory. PyFory's ReduceSerializer could bypass documented DeserializationPolicy validation hooks during reduce-state restoration and global-name resolution. An application is vulnerable if it deserializes attacker-controlled data using PyFory Python-native mode with strict mode disabled and relies on DeserializationPolicy to restrict unsafe classes, functions, or module attributes. This issue affects Apache Fory: from before 1.0.0. Mitigation: Users of Apache Fory are recommended to upgrade to version 1.0.0 or later, which enforces DeserializationPolicy validation for the affected ReduceSerializer paths and thus fixes this issue.

  9. CVE-2026-46556 6.5 Moyenne

    flaskbb

    FlaskBB is a Forum Software written in Python using the micro framework Flask. Prior to version 2.2.1, a Server-Side Request Forgery (SSRF) vulnerability in get_image_info() allows any authenticated user to force the server to send HTTP requests to arbitrary internal endpoints, including cloud metadata services. This is a blind SSRF with confirmed internal port scanning and internal API triggering capabilities. Version 2.2.1 patches the issue.

  10. CVE-2026-47391 9.8 Critique

    MervinPraison PraisonAI

    PraisonAI is a multi-agent teams system. Prior to version 4.6.40, PraisonAI's first-party A2A server example exposes an unauthenticated A2A JSON-RPC endpoint and registers a `calculate(expression)` tool implemented with Python `eval()`. The example also binds to `0.0.0.0`. A remote unauthenticated attacker can send `message/send` to `/a2a`; the request reaches `agent.chat()`, and a real LLM can invoke the registered `calculate` tool. In testing with `gemini/gemini-2.5-flash-lite`, this resulted in arbitrary Python execution in the server process, confirmed by creation of a marker file from an unauthenticated HTTP request. The issue affects deployments following the official A2A example or similar unauthenticated public A2A deployments with unsafe tools. The default unauthenticated A2A surface also exposes task history and task cancellation APIs, increasing confidentiality and integrity impact. Version 4.6.40 patches the issue.

  11. CVE-2026-44190 7.8 Élevée

    Red Hat Ansible Automation Platform 2

    A flaw was found in the Ansible Lightspeed Visual Studio Code extension. This Command Injection vulnerability (CWE-78) allows a remote attacker to execute unauthorized commands on a user's system. The issue occurs because the `ansible.python.activationScript` setting, intended for a virtual environment activation script, does not properly validate user input as a file path. If a user opens or executes a specially crafted project, an attacker could exploit this to gain complete control over the user's system with the privileges of the Visual Studio Code application.

  12. CVE-2026-9291 7.1 Élevée

    AWS Amazon Braket Python SDK

    Insecure deserialization in the job results processing component in Amazon Braket SDK before 1.117.0 might allow a remote authenticated user with S3 write access to the job output bucket to achieve arbitrary code execution on any machine that processes job results. We recommend you upgrade to amazon-braket-sdk version 1.117.0 or later.

  13. CVE-2026-5843 8.6 Élevée

    Docker Desktop · Docker docker desktop

    The MLX inference backend in Docker Model Runner on macOS uses the MLX-LM library, which unconditionally imports and executes arbitrary Python files from model directories via the model_file configuration field in config.json. When a model's config.json specifies a model_file pointing to a Python file, MLX-LM uses importlib to load and execute it with no trust_remote_code gate or equivalent safety check. The MLX backend runs without sandboxing, resulting in arbitrary code execution on the Docker host as the Docker Desktop user. Any container on the Docker network can trigger this by calling the model-runner.docker.internal API to pull a malicious model from an attacker-controlled OCI registry and request inference.

  14. CVE-2026-5817 8.6 Élevée

    Docker Desktop · Docker docker desktop

    The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model tokenizers, and runs without sandboxing. This causes transformers.AutoTokenizer.from_pretrained() to import and execute arbitrary Python files included in any model pulled from an OCI registry, resulting in arbitrary code execution on the Docker host as the Docker Desktop user when inference is triggered. Any container on the Docker network can trigger this by calling the model-runner.docker.internal API to pull a malicious model and request inference.

  15. CVE-2026-47734 5.7 Moyenne

    jelmer dulwich

    Dulwich is a pure-Python implementation of the Git file formats and protocols. Starting in version 0.1.0 and prior to version 1.2.5, a client with push access could push a tiny crafted thin pack (~174 bytes) whose delta header declares a huge dest_size. When dulwich ingested it via add_thin_pack / apply_delta, it would allocate hundreds of MB of memory based on that attacker-controlled size, with no relationship to the actual bytes received. Operators running a Dulwich-based Git server that exposes git-receive-pack (i.e. accepts pushes) - for example via dulwich.server functionality, the HTTP smart server, or anything built on ReceivePackHandler - are impacted. The issue is patched in 1.2.5. add_thin_pack now accepts a max_input_size keyword (bytes; 0/None = unlimited, matching git's semantics), and ReceivePackHandler reads receive.maxInputSize from the repository config and passes it through. Wire reads are counted and a PackInputTooLarge exception is raised once the cap is exceeded - equivalent to git index-pack --max-input-size. Users should upgrade to Dulwich 1.2.5 or later and set receive.maxInputSize in their server's repository config to a sane bound for their environment. On unpatched versions, receive.maxInputSize has no effect, so it cannot be used as a workaround. Until upgrading, operators should restrict dulwich-receive-pack (push) access to trusted, authenticated clients only, or disable it entirely on servers that only need to serve fetches and/or run the server under an OS-level memory limit (e.g. ulimit, cgroups/MemoryMax, or a container memory limit) so a malicious push is killed rather than taking down the host.

  16. CVE-2026-42305 8.8 Élevée

    jelmer dulwich

    Dulwich is a pure-Python implementation of the Git file formats and protocols. Versions starting with 0.10.0 and prior to 1.2.5 have an arbitrary file write leading to remote code execution when cloning or checking out a malicious Git repository on Windows. Dulwich's path-element validator accepted tree entries whose filenames contained bytes that Windows interprets as structural path syntax. Contributing configuration bugs made matters worse. The core.protectNTFS and core.protectHFS settings were looked up under a wrong option name and so user-set values were silently ignored, and core.protectNTFS only defaulted to true on Windows (Git upstream has defaulted it to true everywhere since CVE-2019-1353). Both have been corrected. Anyone who clones, fetches, or checks out an untrusted repository with Dulwich on Windows - either through the Dulwich CLI, porcelain.clone, or any downstream tool built on Dulwich - is impacted. POSIX clones are not directly exploitable (on POSIX \ is a literal filename byte), but a POSIX user can unknowingly propagate a malicious tree to Windows consumers via push or re-publication. This issue is fixed in Dulwich 1.2.5. Users should upgrade to 1.2.5 or later. There is no effective pre-patch workaround. On affected versions the core.protectNTFS configuration key was silently ignored, so setting it to true does not mitigate the issue. Users who cannot upgrade should avoid cloning, fetching, or checking out untrusted repositories with Dulwich on Windows. After upgrading the NTFS validator is on by default on every platform, so no additional configuration is required.

  17. CVE-2026-11393 9.0 Critique

    AWS AgentCore CLI

    Improper neutralization of triple-quote characters during Python code generation in AgentCore CLI before v0.14.2 might allow an authenticated remote threat actor to execute arbitrary code on AWS AgentCore Runtime under the imported agent's IAM execution role and on the local environment of another user in the same AWS account, via a crafted collaborationInstruction stored on a Bedrock Agent collaborator and later processed by that other user during agent import. To remediate this issue, users should upgrade to version 0.14.2.

  18. CVE-2026-45409 5.3 Moyenne

    kjd idna · Kjd internationalized domain names in applications

    Internationalized Domain Names in Applications (IDNA) for Python provides support for Internationalized Domain Names in Applications (IDNA) and Unicode IDNA Compatibility Processing. In versions prior to 3.15, payloads such as `"\u0660" * N` or `"\u30fb" * N + "\u6f22"` utilize the `valid_contexto` function prior to length rejection, and for high values of `N` will take a long time to process. This is the same issue as CVE-2024-3651, however the original remediation in 2024 was not a complete fix. A specially crafted argument to the `idna.encode()` function could consume significant resources. This may lead to a denial-of-service. Starting in version 3.14, the function rejects long inputs as soon as practicable prior to any further processing to minimize resource consumption. In version 3.15, this approach was extended to lesser used alternate functions (i.e. per-label conversions and codec support). A workaround is available. Domain names cannot exceed 253 characters in length. If this length limit is enforced prior to passing the domain to the `idna.encode()` function, it should no longer consume significant resources. This is triggered by arbitrarily large inputs that would not occur in normal usage, but may be passed to the library assuming there is no preliminary input validation by the higher-level application.

  19. CVE-2026-45758 9.6 Critique

    guardrails-ai guardrails · Guardrailsai guardrails ai

    Guardrails AI is a Python framework that helps build AI applications. On May 11, 2026 at approximately 6:00 PM Pacific, an attacker published a malicious version of `guardrails-ai` (0.10.1) to PyPI. Aany user who installed `guardrails-ai==0.10.1` from PyPI on May 11, 2026 may be affected. Security researchers identified the malicious package within approximately 2 hours of publication, and PyPI quarantined the repository. Based on our telemetry, Guardrails AI maintainers have observed no requests to Guardrails AI infrastructure originating from the malicious 0.10.1 version, and a review of system and access logs has produced no evidence of user data exfiltration through their systems. Users should upgrade to version 0.10.2 or downgrade to version 0.10.0, both of which are unaffected. Those who installed version 0.10.1 should rotate any credentials accessible from their machine (GitHub PATs, cloud provider keys, package registry tokens, API keys) and audit their GitHub account for unauthorized workflows or repositories.

  20. CVE-2026-8481 9.9 Critique

    IBM Langflow OSS · Langflow langflow

    IBM Langflow OSS 1.0.0 through 1.10.0 contain a critical remote code execution vulnerability in the code validation API endpoint. The POST /api/v1/validate/code endpoint accepts user-supplied Python code and executes it directly using Python's built-in exec() function without sandboxing, input validation, or privilege restrictions, enabling any authenticated user to execute arbitrary system commands with the full privileges of the Langflow server process.

  21. CVE-2026-8476 9.9 Critique

    IBM Langflow OSS · Langflow langflow

    IBM Langflow OSS 1.0.0 through 1.10.0 contain a critical remote code execution vulnerability in the disk-based caching mechanism. The AsyncDiskCache class uses Python's unsafe pickle.loads() function to deserialize cached objects from disk without validation, integrity verification, or authentication, enabling arbitrary code execution when malicious pickle payloads are processed. Attackers who can influence cached data through file system access, malicious workflow inputs, custom components, or API manipulation can achieve complete system compromise with the privileges of the Langflow server process.

  22. CVE-2026-9135 9.9 Critique

    IBM Langflow OSS · Langflow langflow

    IBM Langflow OSS 1.0.0 through 1.10.0 Langflow versions up to 1.9.2 (commit 94981c443d4918517b9e8163d70fc598dc33a32d) contain a code injection vulnerability in the Policies component's ToolGuard integration that bypasses the allow_custom_components=false security control. The vulnerability exists because the validation mechanism only checks the main component source code in node_template["code"]["value"] but fails to validate dynamic CodeInput fields that store generated ToolGuard Python files. Attackers can embed malicious Python code in these unvalidated dynamic fields, which are persisted in Flow.data and later executed server-side when a guarded tool is invoked through the ToolGuard runtime. This allows authenticated users with flow creation privileges to achieve arbitrary Python code execution on the backend despite custom component restrictions. The vulnerability can be escalated through cross-tenant flow manipulation via the agentic MCP update_flow_component_field tool, which accepts attacker-controlled user_id parameters, enabling attackers to inject malicious code into victim users' flows. When combined with publicly accessible flows and specific misconfigurations (AUTO_LOGIN=true, NEW_USER_IS_ACTIVE=true), the attack can be conducted with reduced authentication requirements.

  23. CVE-2026-14499 8.8 Élevée

    IBM Langflow OSS · Langflow langflow

    IBM Langflow OSS 1.0.0 through 1.10.1 Langflow could allow an authenticated user to execute arbitrary commands with elevated privileges on the system due to improper validation of user supplied input in the Python Interpreter component.