VNC and Intel think offline AI could be the next big thing

VNClagoon and Intel have announced their plans to deliver generative AI to users’ PCs.

With AI taking over more aspects of our lives, the question of how secure it is regarding data remains.

With this in mind, VNC, or Virtual Network Consult AG – an open-source developer for enterprise applications – and Intel have announced that they’re working on generative AI for personal computers.

One of the main concerns when using AI is ensuring data security and privacy.

Large language models (LLMs) are usually hosted by AI vendors on proprietary cloud platforms that process and use data to further develop those models, which may pose risks for security-related applications. For this reason, the two companies provide an alternative that would allow users to use AI while knowing that their data is secure and controlled locally.

To make this work, the companies combined VNClagoon and OpenVINO software toolkit with Intel’s Lunar Lake processors, so that LLMs can be “compressed” and work locally on Intel-driven devices “in a secure sandbox environment.”

This combination allows AI queries to be performed locally, with all data, documents, and models remaining private.

The new system paves the way for “Confidential AI” that provides full functionality and prevents any “unauthorized access” to users’ data. It also allows institutions such as aid organizations, the police, and military to use models in local and secure enviroments.

It has been previously reported that major companies have been using publicly available data to train their AI models. What’s alarming is that, according to research, such practices could lead to a “model collapse” once all the data used to train AI is “devoured,” showing that the tool poses not only data-security-related risks but also issues related to content quality.