**Popular AI Library Hacked: Flaw Allowed Attacker-Controlled Models to Run Code**
A critical vulnerability in the Hugging Face Transformers library, a widely-used tool for building artificial intelligence models, has been discovered by security researchers at Pluto Security Inc. The flaw, which allowed attacker-controlled AI models to run arbitrary code on a victim’s machine, was found to occur when a user loaded a model from a malicious source.
The vulnerability, a remote code execution (RCE) bug, was discovered in the popular Transformers library, a tool used by millions of developers to build, train, and deploy AI models. The library, widely adopted in industries such as natural language processing, computer vision, and speech recognition, is a crucial component of many AI applications.
The attack vector, according to Pluto Security, occurs when a user loads a model from a malicious source, either by downloading it from a compromised repository or by receiving a model from an attacker. Once the model is loaded, the attacker-controlled code is executed, potentially leading to unauthorized access, data theft, or other malicious activities.
**What this means for developers**
Developers using the Hugging Face Transformers library should treat this vulnerability as a high-priority issue. To mitigate the risk, they should ensure that they are only loading models from trusted sources, and should be cautious when loading models from third-party repositories or receiving models from unknown sources.
**The Severity of the Flaw**
The vulnerability is rated as critical, as it allows an attacker to execute arbitrary code on a victim’s machine. Pluto Security notes that the attack vector is relatively simple to exploit, making it a significant concern for developers and organizations that rely on the Hugging Face Transformers library.
**Hugging Face Response**
Hugging Face, the company behind the Transformers library, has not yet released a statement on the vulnerability. However, the company is expected to release a patch or update to address the issue in the near future.
In the meantime, developers should exercise caution when using the Hugging Face Transformers library, and should consider alternative libraries or tools that do not suffer from this critical vulnerability.
