Friday 28 March 2025
A team of researchers has developed a new way to keep your personal data safe when using graph neural networks, a type of artificial intelligence that’s increasingly being used in many areas of life. Graphs are like maps that connect dots, and neural networks are computer programs that can learn from patterns in those connections.
The problem is that these graphs often contain sensitive information, such as who your friends are or what websites you visit. Right now, the way to keep this data safe is by encrypting it before sending it over the internet. But encryption has its limits – for example, if someone gets their hands on your encrypted data, they might still be able to figure out what’s inside.
The researchers have come up with a new solution that uses something called Trusted Execution Environments (TEEs) to keep your data safe. TEEs are like secure containers that can run programs without letting them access the rest of your computer or phone. The team has developed a system called GNNVault that uses these TEEs to keep your graph neural networks safe.
GNNVault works by dividing the program into two parts: a public part that runs on your device, and a private part that runs in the secure container. The public part is like a proxy – it talks to the outside world and sends back information, but doesn’t have access to the sensitive data. The private part does all the heavy lifting, using the graph neural network to analyze the data and make decisions.
The team tested their system on several different types of graphs and found that it was able to keep the sensitive information safe even when an attacker tried to steal it. They also compared their system to other methods for keeping data safe and found that GNNVault was much more effective.
One of the biggest challenges the team faced was figuring out how to get the graph neural network to work in the secure container. Graphs are complex and require a lot of computing power, but the TEE has limited resources. The researchers had to come up with creative solutions to make it all fit together.
The potential applications of GNNVault are huge. With this technology, people could use graph neural networks for things like personalized medicine or social network analysis without worrying about their data being stolen. It could also be used in industries where security is a major concern, such as finance or government.
Overall, the team’s work on GNNVault has opened up new possibilities for using graph neural networks in a secure way.
Cite this article: “Keeping Graph Neural Networks Safe with GNNVault”, The Science Archive, 2025.
Graph Neural Networks, Artificial Intelligence, Data Security, Encryption, Trusted Execution Environments, Secure Containers, Graph Analysis, Personalized Medicine, Social Network Analysis, Cybersecurity
