The joys and sorrows of artificial intelligence. Will the new AI-powered personal computers now appearing on shelves and e-commerce sites really be more powerful and secure? More powerful, certainly. But not much more secure. Will embedded AI really simplify our daily lives? Absolutely not, warns a growing number of independent computer analysts. danger Which can be defused, perhaps by resorting to the same predictive intelligence? Not necessarily. Because the danger is already under our keyboards. Thanks, or rather, because of, the combination of latest-generation PCs and new software updates.
Yes, because the main defendants are personal computers with the new "AI Ready" processors that contain a substantial layer of instructions specific to artificial intelligence, ready to be combined with software that unleashes their full potential. Let's talk first of all about the most recent editions of the browser, the programs that allow us to navigate the Web. Everything is already among us. Everything is ready to amplify the well-known evils of the Internet: computer viruses, cyber robbery of sensitive data, the systematic profiling of our existence, identity theft and all the scams that result from it.
The arrival of rogue Internet browsers
The new evolution of artificial intelligence heralds wonders. From Answer Engine that with ChatGPT or Gemini are able to provide us with immediate and direct answers we move on to Agentic AI browsers, which are much more than a powerful Internet browser. Here's Perplexity Comet, the Atlas evolution of ChatGPT, Opera Neon, and very soon the vitaminized offspring of the well-known Chrome or Firefox. Here, they don't just answer our questions. They operate under the banner of predictive algorithms that are immediately applied. Powerful all-purpose assistants, they say. Or, rather, seasoned cheaters.
explains the super IT expert Andrea Ferrario on Tom's Hardware Italia: until now we have been relatively calm because the classic browsers had perfected and applied a cardinal principle, that of segregationSimply put, it's the ability to isolate each individual website visit in a sort of bubble—an isolated portion of our PC's memory—to protect us from viruses and intrusions. This prevents malicious code on a site from accessing other active memory locations containing sensitive data, such as our bank login details.
Segregation in new browsers risks disappearing precisely to equip new navigation software with all the potential by aggregating sensitive data to carry out their task, combining predictive intelligence with the ability to package all the planning operations that we expect from it.
A real danger already today. Ferrario cites the case of Opera Neon: "Every time the AI reads a page for you, the content, along with your data and instructions, is packaged and sent to the cloud server to be processed by the LLMs." The problem is that Large Language Models, powerful algorithms that sweep the entire contents of our PC do not possess "a moral compass nor a technical understanding of the concept of computer security" with the intrinsic inability, precisely because they are dedicated to scanning and assembling all the information that concerns us, to understand if the product of all this is something we really want or if the procedure is the one that is allowing a cyber attack or thehidden installation of a malware.
From the industry many promises and few guarantees
Is the hardware and software industry looking for countermeasures that can reconcile good protection with the power of all this? Yes, but "it's a titanic task if not impossible." Meanwhile, "entrusting control of the browser, the gateway to our digital life, to an intrinsically uncertain system is a very high-risk gamble," warns Ferrario.
Caution would seem warranted. Yet it is precisely the IT industry, with its hunger for profit that drives innovation, that is encouraging us to leap to these systems without first addressing the security issues, warn, among others, the super experts at Kaspersky, one of the world's most renowned antivirus software companies. There is the far-from-remote possibility of exposing the artificial intelligence servers that collect our data todirect intrusion of third parties, starting with cyber criminals.
“This is not a science fiction scenario: in 2023 – we read in the blog Kaspersky – ChatGPT mistakenly revealed fragments of users' chats, and the “share chat” feature, available to users until July 31, 2025, resulted in tens of thousands of conversations being indexed by search engines. available to anyone".
What's more, the new AI browsers aren't just spies. They're even an open door to crime. Kaspersky's blog mentions a disturbing experiment: "Researchers have managed to deceive an agent artificial intelligence within the Comet browser, tricking it into downloading malware onto the owner's computer. They achieved this by sending a fake email to the victim's account, which people could access, falsely claiming it contained blood test results, via a link that downloaded a malicious file to the computer. It's scary.
A tough challenge for the next Windows 12 too
Microsoft strategists are well aware of these problems. And they're trying to catch the snake by the tail. How? The next operating system, Windows 12, hasn't yet been officially unveiled, and it's unknown when it will arrive. Perhaps as early as the end of this year, when the first year of extended support for the still-widespread Windows 10 expires.
With Windows 12, the aim is to fully adopt the so-called Core PC, separating all operating system functions into independent layers, thus taking the concept of segregation that has so far sought to protect us to a higher level. This complete modularity at the operating system level should simplify the arduous task of working together with the AI and software industries to limit the impact we've discussed so far.
The problem is that a different interpretation is circulating among analysts. Microsoft is thus seeking to directly appropriate the models, functions, and related business of artificial intelligence applied to our personal computers. In this way, problem would occur again exactly in the upper layer of our personal computer and would not be solved at all. Will the next few months be decisive? We fervently hope so. In the meantime, before making the leap into the world of AI-PCs, perhaps by installing the new super-intelligent browsers, it is worth adopting a due caution.
