Lundi 24 août 2026

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.