{"description":"Trending threats, MITRE ATT\u0026CK coverage, and detection metadata. Fed continuously.","feed_url":"https://feed.craftedsignal.io/products/sentence-transformers/feed.json","home_page_url":"https://feed.craftedsignal.io/","items":[{"_cs_actors":[],"_cs_cpes":[],"_cs_cves":[{"cvss":9.8,"id":"CVE-2026-68770"}],"_cs_exploited":false,"_cs_has_poc":false,"_cs_poc_references":[],"_cs_products":["sentence-transformers"],"_cs_severities":["critical"],"_cs_tags":[],"_cs_type":"advisory","_cs_vendors":["Hugging Face"],"content_html":"\u003cp\u003eThe sentence-transformers library contains a security control bypass vulnerability (CVE-2026-68770) that allows attackers to achieve arbitrary code execution. The vulnerability resides in the \u003ccode\u003eimport_module_class\u003c/code\u003e helper function within \u003ccode\u003esentence_transformers/util/misc.py\u003c/code\u003e. A flawed guard condition includes an \u003ccode\u003eor os.path.exists(model_name_or_path)\u003c/code\u003e clause, which forces the trust gate to pass whenever the provided path exists on the local filesystem, effectively ignoring the \u003ccode\u003etrust_remote_code=False\u003c/code\u003e argument.\u003c/p\u003e\n\u003cp\u003eAn attacker who can influence the local filesystem or place files in a directory that a target application subsequently loads can trigger this vulnerability. By placing malicious Python files, such as \u003ccode\u003emodeling_*.py\u003c/code\u003e, and referencing them via a \u003ccode\u003emodules.json\u003c/code\u003e file within the model directory, the attacker ensures that the malicious code executes automatically during the model initialization process. This bypasses the intended security contract, leading to code execution within the context of the process loading the SentenceTransformer model.\u003c/p\u003e\n\u003ch2 id=\"impact\"\u003eImpact\u003c/h2\u003e\n\u003cp\u003eSuccessful exploitation results in arbitrary code execution within the process space of any application utilizing the vulnerable sentence-transformers library to load models from untrusted or attacker-influenced local directories. This poses a significant risk to machine learning pipelines, inference servers, and any system that processes externally sourced model data.\u003c/p\u003e\n\u003ch2 id=\"recommendation\"\u003eRecommendation\u003c/h2\u003e\n\u003col\u003e\n\u003cli\u003eAudit applications using \u003ccode\u003esentence-transformers\u003c/code\u003e to identify locations where models are loaded from directories potentially accessible to unauthorized users.\u003c/li\u003e\n\u003cli\u003eImplement strict filesystem permissions to ensure that model directories cannot be modified by non-privileged accounts.\u003c/li\u003e\n\u003cli\u003ePatch the \u003ccode\u003esentence-transformers\u003c/code\u003e library to a version where the \u003ccode\u003eimport_module_class\u003c/code\u003e logic flaw is remediated.\u003c/li\u003e\n\u003cli\u003eMonitor application logs for unexpected subprocess execution or file system access originating from model-loading service accounts.\u003c/li\u003e\n\u003c/ol\u003e\n","date_modified":"2026-07-31T21:46:26Z","date_published":"2026-07-31T21:46:26Z","id":"https://feed.craftedsignal.io/briefs/2026-07-sentence-transformers-rce/","summary":"A security control bypass vulnerability in sentence-transformers (CVE-2026-68770) allows arbitrary code execution during model loading by exploiting a logic flaw in the import_module_class helper.","title":"Arbitrary Code Execution in sentence-transformers via Logic Flaw","url":"https://feed.craftedsignal.io/briefs/2026-07-sentence-transformers-rce/"}],"language":"en","title":"CraftedSignal Threat Feed - Sentence-Transformers","version":"https://jsonfeed.org/version/1.1"}