Man
Professional
- Messages
- 3,077
- Reaction score
- 614
- Points
- 113
Thousands of AI tools with sensitive data have become publicly available.
Thousands of machine learning tools, including those developed by big tech companies, have been made available online, allowing anyone to interact with them and potentially expose sensitive data. A security researcher shared his findings with 404 Media, raising concerns about the safety of such tools.
In his study, Reddit's lead security engineer Charan Akiri noted that the leaks could include not only machine learning models, but also training datasets, hyperparameters, and even raw data used to create the models. According to Akiri, the misconfiguration of such systems allows outsiders to load or run sensitive models and data sets. The engineer stressed that such platforms are intended exclusively for internal use.
Vulnerable tools include MLflow, Kubeflow, and TensorBoard, which are widely used to train and deploy generative AI models in the cloud, as well as to visualize their results. Due to improper configuration, many companies unwittingly provide access to these tools, which can lead to serious leaks.
One such company is Japanese semiconductor manufacturer Renesas Electronics. The researcher managed to find out that one of the tools belongs to Renesas according to the certificates on the control panel. After 404 Media contacted Renesas, the company promptly fixed the problem. However, there were no comments on the situation.
Akiri also noted that his work has touched only a small part of the problem: there are likely many more companies whose tools remain vulnerable, but their owners have not yet been identified.
When the publication had access to several publicly available MLflow samples, they were given the opportunity to create new runs and explore previous user experiments. According to Akiri, there may be about 5000 vulnerable instances of MLflow on the Internet.
Source
Thousands of machine learning tools, including those developed by big tech companies, have been made available online, allowing anyone to interact with them and potentially expose sensitive data. A security researcher shared his findings with 404 Media, raising concerns about the safety of such tools.
In his study, Reddit's lead security engineer Charan Akiri noted that the leaks could include not only machine learning models, but also training datasets, hyperparameters, and even raw data used to create the models. According to Akiri, the misconfiguration of such systems allows outsiders to load or run sensitive models and data sets. The engineer stressed that such platforms are intended exclusively for internal use.
Vulnerable tools include MLflow, Kubeflow, and TensorBoard, which are widely used to train and deploy generative AI models in the cloud, as well as to visualize their results. Due to improper configuration, many companies unwittingly provide access to these tools, which can lead to serious leaks.
One such company is Japanese semiconductor manufacturer Renesas Electronics. The researcher managed to find out that one of the tools belongs to Renesas according to the certificates on the control panel. After 404 Media contacted Renesas, the company promptly fixed the problem. However, there were no comments on the situation.
Akiri also noted that his work has touched only a small part of the problem: there are likely many more companies whose tools remain vulnerable, but their owners have not yet been identified.
When the publication had access to several publicly available MLflow samples, they were given the opportunity to create new runs and explore previous user experiments. According to Akiri, there may be about 5000 vulnerable instances of MLflow on the Internet.
Source
