Semaine du 24 août au 30 août 2026

dataelement bisheng

BISHENG before 2.6.0 contains a remote code execution vulnerability in the workflow run_once endpoint that allows authenticated users to execute arbitrary Python code. Attackers can submit crafted Code node definitions to the POST /api/v1/workflow/run_once endpoint, which executes them with exec() without sandboxing, gaining access to filesystem, credentials, and internal network resources.

scrapy protego

Protego is a pure-Python robots.txt parser with support for modern conventions. Prior to 0.6.2, protego._urlpattern._URLPattern._prepare_pattern_for_regex translates every asterisk in an Allow or Disallow directive into a lazy regular-expression wildcard, so a directive containing many asterisks creates exponential backtracking. After protego.Protego.parse processes a crafted robots.txt file, protego.Protego.can_fetch can spend an attacker-controlled period matching a near-miss URL and deny service to the crawler. The vulnerable path is src/protego/_urlpattern.py in the _URLPattern match logic. This issue is fixed in version 0.6.2.

piccolo-orm piccolo_admin

Piccolo Admin is an admin interface and content management system for Python, built on top of Piccolo. Prior to 1.14.0, piccolo_admin/endpoints.py uses superuser_validators to block PUT, PATCH, DELETE, and POST requests by non-superusers but permits GET requests to configured user and session tables, while piccolo_api/session_auth/tables.py exposes SessionsBase.token because the token column is not secret. In deployments that add the Sessions and User tables to create_admin, a non-superuser administrator can call GET /api/tables/sessions/, obtain another user's live session token, replay it as the Cookie id value to impersonate a superuser, and permanently set superuser to true on the attacker's own row. This issue is fixed in version 1.14.0.

speechbrain

SpeechBrain before 1.1.1 contains an arbitrary code execution vulnerability that allows attackers to execute arbitrary code by supplying a crafted CKPT.yaml checkpoint metadata file parsed with PyYAML's unsafe loader during candidate enumeration in Checkpointer.recover_if_possible(). Attackers can embed malicious Python object construction tags such as !!python/object/apply in any CKPT.yaml file within the configured checkpoint path to trigger code execution during candidate discovery, even if the malicious checkpoint is never selected for recovery.

jahlives openssl_encrypt · Jahlives openssl encrypt

openssl-encrypt (pip package) before 1.4.9 contains a symlink-following flaw in its verify-usb v2 added-file allowlist scan. The scan enumerated the drive with rglob(), which in CPython does not descend into symlinked directories and treats the symlink as an ordinary directory, while O_NOFOLLOW on the hash side binds only the final path component. An evil-maid attacker with physical access to the removable drive could replace a tool-tree directory with a symlink to a copy containing byte-identical files plus a planted __pycache__/*.pyc file (which CPython loads in preference to recompiling the clean .py). The planted file is never enumerated, added_files stays 0, and verify-usb reports PASSED, resulting in code execution when the victim runs the portable install. Fixed in 1.4.9 (affects both 1.4.x and 1.5.x lines).

flowintel

Affected versions of Flowintel improperly trust configuration keys supplied to the alerts settings update endpoint. While configuration values were normalized to Python literals, the corresponding keys were used directly when constructing and replacing lines in conf/config_module.py. The vulnerable code used requester-controlled keys in both the regular expression and the generated assignment: f'{key} = {py_val}' and appended an assignment if the key was not already present. The modified Python configuration module was subsequently reloaded using importlib.reload(). This creates a code-generation boundary in which specially crafted configuration keys can alter the Python source structure and result in execution of attacker-controlled Python statements. Version impacted >=3.3.0

frappe

Frappe is a full-stack web application framework written in Python and JavaScript. Prior to version 15.115.0, an access control bypass in the REST API allows a user to read data from Linked DocTypes that they are not authorized to access. When a document references another document through a Link field, the framework does not consistently enforce the linked DocType's own permissions when the record is retrieved through the REST API, so a low-privileged authenticated user can obtain fields from linked records outside their permitted scope. This issue is fixed in version 15.115.0.

Andyyyy64 whichllm

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.

nltk · Nltk nltk

NLTK versions before 3.10.0 contain a regular expression denial of service vulnerability in Text.findall() and TokenSearcher.findall() methods that accept user-supplied regular expressions without validation or timeout. Attackers can supply crafted regex patterns that cause catastrophic backtracking, resulting in indefinite CPU saturation and denial of service to all users of the Python process.

oscal-compass compliance-trestle

Compliance-trestle (Trestle) is a Python SDK and command-line tool for managing OSCAL compliance documents. In versions before 3.12.4 and versions 4.0.0 through 4.0.3, the catalog-generate, profile-generate, and ssp-generate author commands write generated Markdown to an attacker-influenced output path without path-traversal validation, allowing arbitrary file write outside the Trestle workspace. These commands join the user-supplied output argument onto the Trestle root and write to the result, but guard it only with an is_directory_name_allowed() task-name-collision check rather than the PathSecurityValidator.validate_local_path() guard used by the jinja command, so an absolute path or one containing traversal sequences escapes the workspace and writes files under an attacker-chosen location as the invoking process owner. The security boundary is crossed when a trusted CI job, shared service, or wrapper derives the output argument from repository-controlled, tenant-controlled, or otherwise untrusted data while expecting output to stay inside the workspace. When --force-overwrite is used, the selected output directory is first recursively deleted, extending the primitive to destruction of an attacker-chosen directory tree and enabling indirect code execution by overwriting files a pipeline later runs. This issue is fixed in versions 3.12.4 and 4.1.0.

oscal-compass compliance-trestle

Compliance-trestle (Trestle) is a Python SDK and command-line tool for managing OSCAL compliance documents. In versions prior to 3.12.4 and 4.0.0 through 4.0.3, the custom Jinja2 include tags mdsection_include and md_clean_include re-parse the content of an included Markdown file as Jinja2 template code in a non-sandboxed environment, allowing server-side template injection that can lead to arbitrary code execution. The MDSectionInclude and MDCleanInclude tags in Trestle/core/jinja/tags.py pass included file content to Parser(self.environment, ...).parse(), splicing it into the host template's compilation, and the environment is a plain jinja2.Environment rather than a SandboxedEnvironment, so any expressions in the file are evaluated with full access to the usual SSTI gadget chain. Because Trestle's Markdown writers emit OSCAL prose and component-description fields verbatim, applying delimiter neutralization only to parameter tables, attacker-controlled OSCAL data such as a control statement, part prose, or component description containing Jinja2 syntax flows into an included Markdown file and is executed when the include tag re-parses it. This issue is fixed in version 4.1.0.

oscal-compass compliance-trestle

Compliance-trestle (Trestle) is a Python SDK and command-line tool for managing OSCAL compliance documents. In versions before 3.12.4 and versions 4.0.0 through 4.0.3, Trestle is vulnerable to server-side template injection that can lead to remote code execution. This occurs because the MDCleanInclude and MDSectionInclude Jinja2 tags re-parse untrusted Markdown content as template source code using a non-sandboxed jinja2.Environment. An attacker who controls content that Trestle renders, such as a crafted workspace Markdown file, a third-party SSP document, or a YAML lookup-table value, can inject a Jinja2 expression that traverses Python object internals to execute arbitrary operating system commands in the context of the Trestle process. This issue is fixed in versions 3.12.4 and 4.1.0.

mesop-dev mesop

Mesop is a Python-based UI framework that allows users to build web applications. Prior to 1.3.3, applications running in debug mode expose a GET /hot-reload endpoint whose unbounded loop depends on the user-supplied counter parameter, allowing an unauthenticated attacker to hold worker threads with high counter values until the worker pool is exhausted and the server becomes unavailable. A single unauthenticated attacker can crash the Mesop server with minimal effort. Because the attack leverages worker exhaustion, the server remains unresponsive until it is manually restarted. This issue is fixed in version 1.3.3.

collective icalendar

icalendar is an RFC 5545 compatible parser and generator of iCalendar files for Python. From 7.1.0 until 7.1.3, the Component equality method in src/icalendar/cal/component.py compares nested subcomponents with two membership loops, and each membership test invokes the same method on child components, causing O(2^n) work relative to nesting depth. Component.from_ical accepts arbitrarily nested BEGIN:VEVENT blocks without a depth limit, so an attacker can submit a sub-kilobyte .ics file containing equal nested subtrees and trigger the cost when an application performs equality, inequality, membership, deduplication, test-assertion, round-trip, or normalization comparisons. Parsing alone does not trigger the issue, and comparisons that differ early short-circuit, but a few hundred bytes can pin a CPU core for minutes or indefinitely, causing denial of service in calendar sync or import endpoints, invite processing, and other comparison paths. This issue is fixed in version 7.1.3.

chainlit

Chainlit is a Python framework for building production-ready conversational AI applications. From 2.4.0rc0 until 2.12.0, Chainlit deployments with features.mcp.enabled set to true in .chainlit/config.toml expose the POST /mcp endpoint without requiring authentication. For sse and streamable-http transports, ConnectSseMCPRequest and ConnectStreamableHttpMCPRequest in backend/chainlit/types.py accept a user-controlled url and optional headers dictionary without scheme validation, private-address filtering, or an allowlist. The connect_mcp handler in backend/chainlit/server.py passes these values to sse_client() or streamablehttp_client(), allowing the Chainlit server to make blind outbound requests to arbitrary internal or external services, including cloud metadata endpoints, with attacker-controlled Authorization and Cookie headers. The SSE URL sink has existed since 2.4.0rc0, while attacker-controlled header forwarding and streamable-http support were added in 2.6.4. The response is consumed internally and not returned, but the attacker can issue state-changing authenticated requests, discover internal services, scan ports, and probe metadata endpoints. This issue is fixed in version 2.12.0.

Amazon strands-agents-tools

Improper neutralization of input used for LLM prompting in the python_repl tool in Amazon Strands Agents Tools before 0.8.5 might allow remote actors to execute arbitrary Python code on the agent's host by bypassing the human consent gate, via a crafted prompt that forwards non_interactive_mode as a keyword argument through the batch tool. To remediate this issue, users should upgrade to version 0.8.5 or later.

GOVCERT-LU eml_parser

eml_parser serves as a python module for parsing eml files and returning various information found in the e-mail as well as computed information. Prior to 3.0.2, eml_parser.routing.noparenthesis in eml_parser/routing.py removes parenthesized CFWS comments from Received: headers with a regex-based fix-point loop whose running time is quadratic in the nesting depth. A single Received: header with 5,000 nested parentheses causes approximately 1.3 seconds of CPU saturation per parsed message, and doubling the nesting depth approximately quadruples the running time. An attacker can submit relatively small EML files that consume multiple seconds of processing time, causing worker latency, queue backpressure, and possible service-level outages in synchronous gateways, sandboxes, and real-time triage pipelines. This issue is fixed in version 3.0.2.

QWED-AI qwed-verification

QWED is open-source AI verification infrastructure for deterministic verification of LLM outputs, tool calls, code, schemas, and agent state before production execution. Prior to 5.1.2, the qwed package passes caller-controlled math expressions directly to SymPy parse_expr() without restricted global_dict and local_dict namespaces, allowing Python eval() to resolve builtins and execute arbitrary Python code in the API server process. In src/qwed_new/api/main.py, POST /verify/math is protected by get_current_tenant but accepts any valid tenant API key, reads the expression field, applies only a cosmetic re.sub(r'(\d)(()', r'\1*\2', expression) normalization, and passes the result to parse_expr(). In src/qwed_new/core/batch.py, POST /verify/batch sends math items through batch_service.create_job(), stores item.query verbatim, and _verify_item() passes VerificationType.MATH input to parse_expr() without sanitization. The default-enabled POST /auth/signup endpoint allows anyone to create a standard tenant account, POST /auth/api-keys issues an x-api-key, and either vulnerable path can then be used to read or write files, modify data, execute operating system commands, terminate the service, and compromise other tenants in a shared deployment. This issue is fixed in version 5.1.2.

gitpython-developers GitPython · Gitpython project gitpython

GitPython before 3.1.59 omits --separate-git-dir from unsafe_git_clone_options, allowing attackers to create arbitrary git directories outside the intended clone destination. Attackers can pass a separate_git_dir parameter to Repo.clone_from() or Repo.clone() to redirect repository metadata to an attacker-controlled filesystem path, enabling arbitrary directory creation and potential hook execution.

gitpython-developers GitPython · Gitpython project gitpython

GitPython before 3.1.59 fails to disable merge_includes when parsing .gitmodules, allowing attackers to disclose local file content by including arbitrary file paths via [include] directives. Attackers can craft a malicious .gitmodules file with include directives pointing to sensitive files; when repo.submodules is accessed, GitConfigParser raises MissingSectionHeaderError embedding the target file's first line verbatim in the exception message.

lepture mistune

Mistune is a Python Markdown parser with renderers and plugins. Versions 3.3.0 through 3.3.2 are vulnerable to DoS through deeply nested tokens. HTML rendering creates deeply nested emphasis tokens from consecutive asterisk characters, and recursive rendering in HTMLRenderer.render_token() can exceed Python's recursion limit and raise RecursionError, allowing crafted Markdown to crash a parsing process. This issue is fixed in version 3.3.3

flairNLP flair

The official Flair wheels for 0.15.0 and 0.15.1 still contain flair/models/clustering.py, whose ClusteringModel.load static method returns pickle.loads(joblib.load(str(model_file))) and so executes arbitrary Python while loading a model file. Loading a model supplied by an attacker therefore runs that attacker's code with the privileges of the loading process. This is the same sink and the same file as CVE-2024-10073, which records 0.15.0 as the fixed version on the basis that clustering support was dropped in that release; the module was removed from the documented API but remains present in the distributed artifact and reachable by importing flair.models.clustering directly, so the earlier record's fixed version does not hold for the shipped package.

xorbitsai inference

Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. Six loader call sites pass trust_remote_code=True as a literal or as an unconditional default: RerankModel._get_tokenizer in xinference/model/rerank/core.py, SentenceTransformerRerankModel.load in xinference/model/rerank/sentence_transformers/core.py, SentenceTransformerEmbeddingModel.load in xinference/model/embedding/sentence_transformers/core.py, FlagEmbeddingModel.load in xinference/model/embedding/flag/core.py, and two sites in xinference/model/llm/transformers/core.py where PytorchModel._sanitize_model_config and PytorchModel._get_components default the value to True. Because a caller with model launch access can register a model whose type is unknown and supply an arbitrary model path, the server reaches _auto_detect_type and then AutoTokenizer.from_pretrained, which imports and executes Python declared by the model directory's own tokenizer_config.json auto_map, running attacker-supplied code with the privileges of the worker process. Version 2.12.0 gates every site behind allow_trust_remote_code and the XINFERENCE_TRUST_REMOTE_CODE setting, permitting remote code only for bundled built-in models.