We likely don’t realize how much AI plays a role in our daily lives. You know those annoying phone trees, the ones that don’t seem to ask the right questions for your problem or offer a good answer or access to a real human being? That’s AI at work. They drive me nuts, but what keeps me awake is who has access to my data?

Many of our customers are turning to Microsoft Copilot to access the power of AI, and if you’re thinking about doing it, here are a couple of things to consider.

As you’ll discover upfront, there’s a free version and a paid version. The main difference is that free Copilot is a basic AI assistant with web-grounded chat and limited image creation. The paid Copilot Pro and Microsoft 365 Copilot offer deeper integration with Microsoft apps, priority access to advanced models, and higher usage limits. I can use ChatGPT to create Excel formulas for my data, but it’s the paid version that accesses my data.

AI carries a number of risks, including data poisoning, adversarial attacks, and privacy leakage, which can compromise a model’s integrity and sensitive data. There’s also the potential for model theft and vulnerabilities in the supply chain and APIs. Let’s focus on risks related to data, privacy, and model integrity.

• Privacy Leakage: AI models trained on sensitive data may inadvertently leak that information through their outputs. This includes data inversion and membership inference attacks, where attackers try to extract private information about the training data.
• Model Stealing: Attackers can reverse-engineer or replicate an AI model by analyzing its outputs, which can be used for malicious purposes or to steal intellectual property.
• Data Breaches: AI systems often require large amounts of data, making them attractive targets for data theft. A breach can expose sensitive personal, financial, or proprietary information.

Whether you use AI or not, Windows 11 and your computer play key roles in your security. All computers are not created equal. If you do a lot of work with Copilot, your computer may not cut it. You should have a computer with a neural processing unit (NPU) capable of processing 40 TOPS – 40 trillion operations per second. Anything less than that will require your computer to offload data from your CPU and graphics card by sending it to the cloud.

Sending it to the cloud involves a security risk, no matter how small the risk may be, and that’s a breach opportunity. Sending data to the cloud also slows you down. If your computer can keep all your work local, it’s faster and more private.

Windows-based computer chips that run 40+ TOPS or more are the specialized Neural Processing Units (NPUs) in new “Copilot+ PCs,” which include processors from Intel’s Lunar Lake series and upcoming Qualcomm Snapdragon X Elite/Plus chips. These are not standard CPUs like the Intel Core Ultra 7 155H, which does not meet the requirement.

Computers capable of 40 TOPS start at around $600 to $700. More powerful and versatile models can cost more than $1,000, but prices could be much higher, depending on the NPU or if it uses a more expensive, high-performance GPU, which can add significant cost for graphics-intensive tasks. In addition, you likely will have licensing fees depending on what you’re doing and how many computers are doing the work.

We can help you assess your AI needs and sort through myriad options for Copilot licenses and the computers needed to accomplish the tasks you require. AI can require a large investment, which requires intensive investigation. Contact us by phone – 973-433-6676 – or email to set up an appointment to start the investigation process.