CVE-2026-68770 is a critical security control bypass vulnerability in the sentence-transformers library by Hugging Face. The vulnerability arises from a logic flaw in the import_module_class helper within sentence_transformers/util/misc.py. The guard condition improperly trusts any existing local path due to an 'or os.path.exists(model_name_or_path)' clause, ignoring the trust_remote_code=False argument. Consequently, attackers who can control or influence the contents of a model directory on disk can place malicious Python files (e.g., modeling_*.py referenced via modules.json). These files execute at import time when the application loads the model with SentenceTransformer(path, trust_remote_code=False), leading to arbitrary code execution and bypassing the documented security contract.

An attacker able to control or influence the contents of a model directory on disk can achieve arbitrary code execution within the context of the application loading the model. This bypasses the trust_remote_code=False security setting, potentially allowing full compromise of the host running the vulnerable sentence-transformers library.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, avoid loading models from untrusted or user-controlled directories. Do not rely solely on the trust_remote_code=False argument to prevent code execution from local model paths.