NLTK (Natural Language Toolkit) is a suite of open source Python modules, data sets, and tutorials supporting research and development in Natural Language Processing. Versions prior to 3.6.5 are vulnerable to regular expression denial of service (ReDoS) attacks. The vulnerability is present in PunktSentenceTokenizer, sent_tokenize and word_tokenize. Any users of this class, or these two functions, are vulnerable to the ReDoS attack. In short, a specifically crafted long input to any of these vulnerable functions will cause them to take a significant amount of execution time. If your program relies on any of the vulnerable functions for tokenizing unpredictable user input, then we would strongly recommend upgrading to a version of NLTK without the vulnerability. For users unable to upgrade the execution time can be bounded by limiting the maximum length of an input to any of the vulnerable functions. Our recommendation is to implement such a limit.

A Buffer Overflow vulnerability exists in NumPy 1.9.x in the PyArray_NewFromDescr_int function of ctors.c when specifying arrays of large dimensions (over 32) from Python code, which could let a malicious user cause a Denial of Service. NOTE: The vendor does not agree this is a vulneraility; In (very limited) circumstances a user may be able provoke the buffer overflow, the user is most likely already privileged to at least provoke denial of service by exhausting memory. Triggering this further requires the use of uncommon API (complicated structured dtypes), which is very unlikely to be available to an unprivileged user

vault-cli is a configurable command-line interface tool (and python library) to interact with Hashicorp Vault. In versions before 3.0.0 vault-cli features the ability for rendering templated values. When a secret starts with the prefix `!template!`, vault-cli interprets the rest of the contents of the secret as a Jinja2 template. Jinja2 is a powerful templating engine and is not designed to safely render arbitrary templates. An attacker controlling a jinja2 template rendered on a machine can trigger arbitrary code, making this a Remote Code Execution (RCE) risk. If the content of the vault can be completely trusted, then this is not a problem. Otherwise, if your threat model includes cases where an attacker can manipulate a secret value read from the vault using vault-cli, then this vulnerability may impact you. In 3.0.0, the code related to interpreting vault templated secrets has been removed entirely. Users are advised to upgrade as soon as possible. For users unable to upgrade a workaround does exist. Using the environment variable `VAULT_CLI_RENDER=false` or the flag `--no-render` (placed between `vault-cli` and the subcommand, e.g. `vault-cli --no-render get-all`) or adding `render: false` to the vault-cli configuration yaml file disables rendering and removes the vulnerability. Using the python library, you can use: `vault_cli.get_client(render=False)` when creating your client to get a client that will not render templated secrets and thus operates securely.

Rapid7 Insight Agent, versions 3.0.1 to 3.1.2.34, suffer from a local privilege escalation due to an uncontrolled DLL search path. Specifically, when Insight Agent versions 3.0.1 to 3.1.2.34 start, the Python interpreter attempts to load python3.dll at "C:\DLLs\python3.dll," which normally is writable by locally authenticated users. Because of this, a malicious local user could use Insight Agent's startup conditions to elevate to SYSTEM privileges. This issue was fixed in Rapid7 Insight Agent 3.1.2.35. This vulnerability is a regression of CVE-2019-5629.

lxml is a library for processing XML and HTML in the Python language. Prior to version 4.6.5, the HTML Cleaner in lxml.html lets certain crafted script content pass through, as well as script content in SVG files embedded using data URIs. Users that employ the HTML cleaner in a security relevant context should upgrade to lxml 4.6.5 to receive a patch. There are no known workarounds available.

Synapse is a package for Matrix homeservers written in Python 3/Twisted. Prior to version 1.47.1, Synapse instances with the media repository enabled can be tricked into downloading a file from a remote server into an arbitrary directory. No authentication is required for the affected endpoint. The last 2 directories and file name of the path are chosen randomly by Synapse and cannot be controlled by an attacker, which limits the impact. Homeservers with the media repository disabled are unaffected. Homeservers with a federation whitelist are also unaffected, since Synapse will check the remote hostname, including the trailing `../`s, against the whitelist. Server administrators should upgrade to 1.47.1 or later. Server administrators using a reverse proxy could, at the expense of losing media functionality, may block the certain endpoints as a workaround. Alternatively, non-containerized deployments can be adapted to use the hardened systemd config.

The AWS IoT Device SDK v2 for Java, Python, C++ and Node.js appends a user supplied Certificate Authority (CA) to the root CAs instead of overriding it on macOS systems. Additionally, SNI validation is also not enabled when the CA has been “overridden”. TLS handshakes will thus succeed if the peer can be verified either from the user-supplied CA or the system’s default trust-store. Attackers with access to a host’s trust stores or are able to compromise a certificate authority already in the host's trust store (note: the attacker must also be able to spoof DNS in this case) may be able to use this issue to bypass CA pinning. An attacker could then spoof the MQTT broker, and either drop traffic and/or respond with the attacker's data, but they would not be able to forward this data on to the MQTT broker because the attacker would still need the user's private keys to authenticate against the MQTT broker. The 'aws_tls_ctx_options_override_default_trust_store_*' function within the aws-c-io submodule has been updated to address this behavior. This issue affects: Amazon Web Services AWS IoT Device SDK v2 for Java versions prior to 1.5.0 on macOS. Amazon Web Services AWS IoT Device SDK v2 for Python versions prior to 1.7.0 on macOS. Amazon Web Services AWS IoT Device SDK v2 for C++ versions prior to 1.14.0 on macOS. Amazon Web Services AWS IoT Device SDK v2 for Node.js versions prior to 1.6.0 on macOS. Amazon Web Services AWS-C-IO 0.10.7 on macOS.

The AWS IoT Device SDK v2 for Java, Python, C++ and Node.js appends a user supplied Certificate Authority (CA) to the root CAs instead of overriding it on Unix systems. TLS handshakes will thus succeed if the peer can be verified either from the user-supplied CA or the system’s default trust-store. Attackers with access to a host’s trust stores or are able to compromise a certificate authority already in the host's trust store (note: the attacker must also be able to spoof DNS in this case) may be able to use this issue to bypass CA pinning. An attacker could then spoof the MQTT broker, and either drop traffic and/or respond with the attacker's data, but they would not be able to forward this data on to the MQTT broker because the attacker would still need the user's private keys to authenticate against the MQTT broker. The 'aws_tls_ctx_options_override_default_trust_store_*' function within the aws-c-io submodule has been updated to override the default trust store. This corrects this issue. This issue affects: Amazon Web Services AWS IoT Device SDK v2 for Java versions prior to 1.5.0 on Linux/Unix. Amazon Web Services AWS IoT Device SDK v2 for Python versions prior to 1.6.1 on Linux/Unix. Amazon Web Services AWS IoT Device SDK v2 for C++ versions prior to 1.12.7 on Linux/Unix. Amazon Web Services AWS IoT Device SDK v2 for Node.js versions prior to 1.5.3 on Linux/Unix. Amazon Web Services AWS-C-IO 0.10.4 on Linux/Unix.

Connections initialized by the AWS IoT Device SDK v2 for Java (versions prior to 1.4.2), Python (versions prior to 1.6.1), C++ (versions prior to 1.12.7) and Node.js (versions prior to 1.5.3) did not verify server certificate hostname during TLS handshake when overriding Certificate Authorities (CA) in their trust stores on MacOS. This issue has been addressed in aws-c-io submodule versions 0.10.5 onward. This issue affects: Amazon Web Services AWS IoT Device SDK v2 for Java versions prior to 1.4.2 on macOS. Amazon Web Services AWS IoT Device SDK v2 for Python versions prior to 1.6.1 on macOS. Amazon Web Services AWS IoT Device SDK v2 for C++ versions prior to 1.12.7 on macOS. Amazon Web Services AWS IoT Device SDK v2 for Node.js versions prior to 1.5.3 on macOS. Amazon Web Services AWS-C-IO 0.10.4 on macOS.

Connections initialized by the AWS IoT Device SDK v2 for Java (versions prior to 1.3.3), Python (versions prior to 1.5.18), C++ (versions prior to 1.12.7) and Node.js (versions prior to 1.5.1) did not verify server certificate hostname during TLS handshake when overriding Certificate Authorities (CA) in their trust stores on Windows. This issue has been addressed in aws-c-io submodule versions 0.9.13 onward. This issue affects: Amazon Web Services AWS IoT Device SDK v2 for Java versions prior to 1.3.3 on Microsoft Windows. Amazon Web Services AWS IoT Device SDK v2 for Python versions prior to 1.5.18 on Microsoft Windows. Amazon Web Services AWS IoT Device SDK v2 for C++ versions prior to 1.12.7 on Microsoft Windows. Amazon Web Services AWS IoT Device SDK v2 for Node.js versions prior to 1.5.3 on Microsoft Windows.

A flaw was found in python-pip in the way it handled Unicode separators in git references. A remote attacker could possibly use this issue to install a different revision on a repository. The highest threat from this vulnerability is to data integrity. This is fixed in python-pip version 21.1.

The verify function in the Stark Bank Python ECDSA library (aka starkbank-escada or ecdsa-python) before 2.0.1 fails to check that the signature is non-zero, which allows attackers to forge signatures on arbitrary messages.

EnroCrypt is a Python module for encryption and hashing. Prior to version 1.1.4, EnroCrypt used the MD5 hashing algorithm in the hashing file. Beginners who are unfamiliar with hashes can face problems as MD5 is considered an insecure hashing algorithm. The vulnerability is patched in v1.1.4 of the product. As a workaround, users can remove the `MD5` hashing function from the file `hashing.py`.

Python discord bot is the community bot for the Python Discord community. In affected versions when a non-blacklisted URL and an otherwise triggering filter token is included in the same message the token filter does not trigger. This means that by including any non-blacklisted URL moderation filters can be bypassed. This issue has been resolved in commit 67390298852513d13e0213870e50fb3cff1424e0

TensorFlow is an open source platform for machine learning. In affected versions the code behind `tf.function` API can be made to deadlock when two `tf.function` decorated Python functions are mutually recursive. This occurs due to using a non-reentrant `Lock` Python object. Loading any model which contains mutually recursive functions is vulnerable. An attacker can cause denial of service by causing users to load such models and calling a recursive `tf.function`, although this is not a frequent scenario. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, as these are also affected and still in supported range.

An issue was discovered in the Dask distributed package before 2021.10.0 for Python. Single machine Dask clusters started with dask.distributed.LocalCluster or dask.distributed.Client (which defaults to using LocalCluster) would mistakenly configure their respective Dask workers to listen on external interfaces (typically with a randomly selected high port) rather than only on localhost. A Dask cluster created using this method (when running on a machine that has an applicable port exposed) could be used by a sophisticated attacker to achieve remote code execution.

Snudown is a reddit-specific fork of the Sundown Markdown parser used by GitHub, with Python integration added. In affected versions snudown was found to be vulnerable to denial of service attacks to its reference table implementation. References written in markdown ` [reference_name]: https://www.example.com` are inserted into a hash table which was found to have a weak hash function, meaning that an attacker can reliably generate a large number of collisions for it. This makes the hash table vulnerable to a hash-collision DoS attack, a type of algorithmic complexity attack. Further the hash table allowed for duplicate entries resulting in long retrieval times. Proofs of concept and further discussion of the hash collision issue are discussed on the snudown GHSA(https://github.com/reddit/snudown/security/advisories/GHSA-6gvv-9q92-w5f6). Users are advised to update to version 1.7.0.

Babel.Locale in Babel before 2.9.1 allows attackers to load arbitrary locale .dat files (containing serialized Python objects) via directory traversal, leading to code execution.

python-tuf is a Python reference implementation of The Update Framework (TUF). In both clients (`tuf/client` and `tuf/ngclient`), there is a path traversal vulnerability that in the worst case can overwrite files ending in `.json` anywhere on the client system on a call to `get_one_valid_targetinfo()`. It occurs because the rolename is used to form the filename, and may contain path traversal characters (ie `../../name.json`). The impact is mitigated by a few facts: It only affects implementations that allow arbitrary rolename selection for delegated targets metadata, The attack requires the ability to A) insert new metadata for the path-traversing role and B) get the role delegated by an existing targets metadata, The written file content is heavily restricted since it needs to be a valid, signed targets file. The file extension is always .json. A fix is available in version 0.19 or newer. There are no workarounds that do not require code changes. Clients can restrict the allowed character set for rolenames, or they can store metadata in files named in a way that is not vulnerable: neither of these approaches is possible without modifying python-tuf.

The bluemonday sanitizer before 1.0.16 for Go, and before 0.0.8 for Python (in pybluemonday), does not properly enforce policies associated with the SELECT, STYLE, and OPTION elements.