Hugging Face Text Embeddings Inference versions up to 1.9.4 have a vulnerability where the configured API key is logged in cleartext due to missing redaction in the router's Args struct. This allows attackers with access to logs, container output, or telemetry data to recover the Bearer token and access protected endpoints.
A vulnerability in Hugging Face Transformers (versions 4.57.0 to 5.16.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). The function fetches and caches a remote module file before performing the required trust_remote_code consent check, inverting the security model enforced by other code-loading paths (such as AutoConfig, AutoModel, and AutoTokenizer). As a result, attacker‑controlled Python code from custom_generate/generate.py is copied into the user’s ~/.cache/huggingface/modules directory even if the user declines the trust prompt. Although execution is correctly gated, the file write is not reversible and can persist across sessions. This can lead to persistent, unauthorized files on disk and stale cache collisions where cached attacker code may later be executed during trusted model loads. The issue stems from an unconditional file write in dynamic_module_utils.py prior to any trust verification.
sentence-transformers contains a security control bypass vulnerability that allows attackers to achieve arbitrary code execution by exploiting a logic flaw in the import_module_class helper within sentence_transformers/util/misc.py, where the guard condition includes an 'or os.path.exists(model_name_or_path)' clause that satisfies the trust gate whenever the supplied path exists on the local filesystem, regardless of the trust_remote_code=False argument. Attackers who can control or influence the contents of a model directory on disk can place malicious Python files such as modeling_*.py referenced via modules.json, causing the code to execute at import time when an application loads the model with SentenceTransformer(path, trust_remote_code=False), bypassing the documented security contract and achieving code execution within the loading process.
LeRobot through 0.5.1 contains an unsafe deserialization vulnerability in the async inference pipeline where pickle.loads() is used to deserialize data received over unauthenticated gRPC channels without TLS in the policy server and robot client components. An unauthenticated network-reachable attacker can achieve arbitrary code execution on the server or client by sending a crafted pickle payload through the SendPolicyInstructions, SendObservations, or GetActions gRPC calls.
Hugging Face Transformers GLM4 Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the parsing of weights. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-28309.
CVE-2025-14929 is a high-severity vulnerability in Hugging Face Transformers affecting the X-CLIP checkpoint conversion process. It arises from improper validation of user-supplied data during deserialization, enabling remote code execution if a user opens a malicious file or visits a crafted webpage. Exploitation requires user interaction but no prior authentication. The vulnerability impacts confidentiality, integrity, and availability by allowing arbitrary code execution within the affected process context. No known exploits are currently in the wild. European organizations using vulnerable versions of Hugging Face Transformers, especially in AI/ML workflows, face significant risk. Mitigation involves avoiding untrusted checkpoint files, applying patches when available, and implementing strict input validation and sandboxing. Countries with strong AI sectors and extensive use of Hugging Face tools, such as Germany, France, the UK, and the Netherlands, are most likely affected.
Hugging Face Transformers HuBERT convert_config Code Injection Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must convert a malicious checkpoint.
The specific flaw exists within the convert_config function. The issue results from the lack of proper validation of a user-supplied string before using it to execute Python code. An attacker can leverage this vulnerability to execute code in the context of the current user. Was ZDI-CAN-28253.
CVE-2025-14927 is a high-severity remote code execution vulnerability in Hugging Face Transformers version 4.57.0, specifically in the convert_config function. It arises from improper validation of user-supplied strings that are executed as Python code, enabling attackers to execute arbitrary code with the privileges of the current user. Exploitation requires user interaction, specifically converting a malicious checkpoint file. The vulnerability impacts confidentiality, integrity, and availability, with a CVSS score of 7.8. There are no known exploits in the wild yet, but the risk is significant due to the widespread use of Hugging Face Transformers in AI and ML workflows. European organizations using this library in development or production environments are at risk, especially those in countries with strong AI sectors. Mitigation involves avoiding untrusted checkpoint files, applying patches once available, and implementing strict input validation and sandboxing.
Hugging Face Accelerate Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Accelerate. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the parsing of checkpoints. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-27985.
Hugging Face Diffusers CogView4 Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Diffusers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the parsing of checkpoints. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-27424.
Hugging Face smolagents Remote Python Executor Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face smolagents. Authentication is not required to exploit this vulnerability.
The specific flaw exists within the parsing of pickle data. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the service account. Was ZDI-CAN-28312.
Hugging Face Transformers megatron_gpt2 Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the parsing of checkpoints. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-27984.
CVE-2025-14920 is a high-severity remote code execution vulnerability in the Hugging Face Transformers library, specifically affecting the Perceiver model deserialization process. The flaw arises from improper validation of user-supplied model files, allowing deserialization of untrusted data. Exploitation requires user interaction, such as opening a malicious file or visiting a crafted webpage. Successful exploitation enables attackers to execute arbitrary code with the privileges of the current user. Although no known exploits are reported in the wild, the vulnerability poses significant risks to confidentiality, integrity, and availability. The vulnerability has a CVSS score of 7.8, reflecting its high impact and relatively low complexity to exploit. European organizations using Hugging Face Transformers in AI/ML workflows, especially those processing untrusted or external model files, are at risk. Mitigation involves strict validation of model inputs, restricting model file sources, and applying patches once available. Countries with strong AI development sectors and extensive use of open-source ML frameworks, such as Germany, France, and the UK, are most likely affected.
CVE-2025-14926 is a high-severity code injection vulnerability in Hugging Face Transformers version 4.57.0, specifically in the convert_config function. It allows remote attackers to execute arbitrary Python code by supplying a malicious checkpoint configuration. Exploitation requires user interaction, as the target must convert a crafted checkpoint file. The vulnerability arises from improper validation of user-supplied strings before execution, enabling code execution with the privileges of the current user. Although no known exploits are currently in the wild, the impact on confidentiality, integrity, and availability is high. This vulnerability poses a significant risk to organizations using Hugging Face Transformers for machine learning model management or deployment. European organizations relying on this library, especially in AI research and development sectors, should prioritize patching or mitigating this issue. Countries with strong AI and tech industries, such as Germany, France, the UK, and the Netherlands, are likely to be most affected.
CVE-2025-14921 is a high-severity remote code execution vulnerability in the Hugging Face Transformers library, specifically affecting the Transformer-XL model deserialization process. The flaw arises from improper validation of user-supplied model files, enabling deserialization of untrusted data. Exploitation requires user interaction, such as visiting a malicious webpage or opening a crafted file, allowing attackers to execute arbitrary code with the privileges of the current user. The vulnerability impacts confidentiality, integrity, and availability of affected systems. No known exploits are currently reported in the wild. European organizations using vulnerable versions of Hugging Face Transformers are at risk, especially those integrating Transformer-XL models in their AI workflows. Mitigation involves strict validation of model files, restricting model sources, and applying updates once patches become available. Countries with significant AI development and adoption, such as Germany, France, and the UK, are more likely to be affected due to higher usage of these tools.
Hugging Face Transformers Trax Model Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the handling of model files. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current user. Was ZDI-CAN-25012.
Hugging Face Transformers MaskFormer Model Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the parsing of model files. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current user. Was ZDI-CAN-25191.
Hugging Face Transformers MobileViTV2 Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face Transformers. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.
The specific flaw exists within the handling of configuration files. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current user. Was ZDI-CAN-24322.
