In my last blog post, I detailed three big risks AI can introduce into our systems. I claimed AI can compromise the confidentiality of our data, throw a wrench in our continuity plans, and take us down the dark road of security through obscurity. However, I ended with a note that open-source, offline AI might help us to dodge these obstacles, and that is the point this post will elaborate on. I will take Jan AI as an example of this kind of solution, since it’s the one I’ve experimented with the most, but know that there are others and the same benefits apply. Jan AI is an open-source platform where AI developers can share their models and make them available to freely download offline. It is a shining example of how offline AI can improve security, redundancy, and the speed of evolution in response to new threats and demands.
Jan AI is a project that has made open-source software allowing you to download various developers’ AI models onto your own device. The first benefit of this is most obvious – privacy! Your data isn’t being piped straight to a company that controls a proprietary model. Instead, it’s staying on your computer along with all your other files. My opinion is this is best practice for everyday use. I use AI everyday to help debug code, make entertainment recommendations, and bounce ideas. It’s not necessary to have cutting edge AI to solve these problems. More than that, these kinds of daily questions can build a very full picture of the types of devices I have, the types of services I use, and the way I speak. This level of detail accumulated over years could pose a grave security risk if it fell into the wrong hands. Not only could it be turned against me, it could be turned against friends, family, or coworkers if someone used that detail to impersonate me and craft an intelligent phishing attack against them. It makes much more sense to offload these kinds of everyday questions to an AI on a device in my personal possession.
Besides privacy, Jan AI adds a critical fail-safe for any project that is using AI for their operations. For example, AI that can replace radiology is on the horizon and I wouldn’t be surprised if a lot of medical institutions go down that road. If they do, they will need extensive back ups and fail-safes in place, and since this route likely involves even fewer human radiologists joining the field, the first fail-safe most places will probably think of is setting up more connections to the internet. The thought process might be to get data flowing in from multiple avenues, so if one goes down, operations can continue uninterrupted. However, this approach directly increases the attack surface. An alternate approach would be utilizing offline AI that can do the same thing. This eliminates the chance of weather issues or network connectivity problems killing operational continuity, while also decreasing the overall attack surface.
Finally, projects like Jan AI avoid the issue of security through obscurity and position us to patch and advance in real time thanks to their open-source foundation. I strongly recommend checking out all the different models available for download on Jan AI. There are diverse philosophies and strategies to building AI, and you can find teams training their models on very different sets of information. You can take inspiration from the code in people’s GitHub repositories and start building your own model. This is something that I would love to do someday. The point is, you will see that passionate, security-minded developers working together leads to far more diverse and interesting outcomes than what the handful of proprietary, cloud-based models on the market can deliver. And if you want to interrogate the security if the model you’re using, you can simply read the code, rather than blindly trusting that your data is being managed well behind closed doors.
Offline AI makes your security posture safer from malicious actors, weatherproof against disasters, and more adaptable to your immediate goals. If you work in Cybersecurity, go check out the free-open source models available on Jan, and consider adding them to your arsenal – Jan is the future!
