Key facts

• NordVPN has launched what it calls next-gen antivirus, or NGAV, a broad security layer for online life rather than a file-scanning tool.
• The company's CTO says the old antivirus model is incomplete because threats now arrive through the internet as phishing, scams and identity theft, often supercharged by AI.
• NordVPN's strategy emphasizes invisible, low-friction protection that runs in the background and does not require technical expertise.
• NordVPN is developing protection for personal AI agents, which it sees as a future attack surface on users' devices.
• The company builds small in-house AI models for tasks such as phishing detection, and those models can run locally on user hardware.
• Internally, NordVPN uses a blend of AI models, increasingly favoring open-weight models as their performance improves and costs matter.
• Briedis says NordVPN does not train on customer data and supports privacy protections, while warning that over-regulation could slow innovation.
• His dream security tool would follow a user's digital exposure, keep data encrypted at rest, in transit and ideally in use, and automatically remove information when needed.

Antivirus meets an AI-shaped threat landscape

Traditional antivirus software was built for a world in which dangerous code arrived as a malicious file: an infected floppy disk, a downloaded executable, a document booby-trapped with a macro. That model produced familiar security rituals, including signature updates, scheduled scans and quarantine folders. But the threat landscape has moved, and the old vocabulary has struggled to keep up. NordVPN's CTO, identified in an interview as Briedis, argues that viruses are no longer primarily arriving through removable media or discrete files. "We don't believe that viruses right now are coming through floppy disks and CDs," he said. "All the protection that you need right now, basically, is on the internet."

That shift is why NordVPN recently launched what it calls next-gen antivirus, or NGAV. The product is positioned as a catch-all security tool for a user's online life, rather than a narrow utility that scans storage drives. The company's move reflects a broader recognition in cybersecurity: file-based attacks still matter, but socially engineered scams, phishing campaigns and identity theft have ballooned into a much larger part of the threat landscape. Those threats are also rapidly scalable with AI, which can help attackers write more convincing messages, clone voices, generate fake websites and automate outreach at a volume that human fraudsters could not easily match.

NordVPN describes NGAV as a security system for a user's entire digital life. It promises to warn users about scam websites, block ads and trackers, and remove malware quietly in the background. The company's first task was securing the VPN tunnel itself. The second, as Briedis framed it, is addressing what comes through that tunnel. That distinction matters because a VPN can protect traffic in transit, but it does not automatically tell a user whether a website is fraudulent, whether an email is a phishing attempt or whether a piece of software is behaving maliciously. NGAV is meant to layer those protections into the same consumer experience.

Why invisible protection is easier to adopt

The most important design choice, according to Briedis, is invisibility. As AI-enabled threats multiply and become more seamless, he believes a solution that runs in the background and does not require technical expertise is the best way to persuade people to adopt better security hygiene. Most users do not want to configure firewall rules, interpret malware warnings or investigate suspicious processes. They want protection that works without turning security into a second job.

That low-friction approach is not just a product preference. It is a response to a stubborn reality: security companies can offer users protection, but they cannot save them from every risky decision. Briedis echoed a point made by leaders at other privacy-focused companies when he described the human element as unavoidable. "There's always a human element," he said. "We usually joke in the IT world, between the geeks, that the weakest link between the computer and the chair is the human."

NordVPN still considers education important. Briedis said the company wants to help people understand the tools they use, where encryption matters, what data they are sharing, with whom they are sharing it and where the risks lie. But education alone is not enough when scams arrive through everyday channels and AI makes them more convincing. A background security layer can reduce the number of decisions users have to get right. It can flag a fraudulent site before a password is entered, block trackers that profile browsing behavior and remove malware without forcing the user to become an amateur incident responder.

AI agents become the next attack surface

The rise of personal AI agents adds another layer of complexity. These agents may operate on a user's device, access personal data, interact with online services and take actions on the user's behalf. That makes them useful, but also vulnerable. An agent could be manipulated, tricked or compromised, and the consequences could extend beyond a single chat session. NordVPN's CTO said the company is looking into how to protect agents themselves, not just the computer or the human operating it.

"We are looking into how to protect agents as well, because we see that agents [themselves] could be attacked on your computer," Briedis said. "I think all cybersecurity is going to go there eventually. It's kind of a no-brainer, because we're going to use [agents] every day." The feature is still in development and does not yet have a name, but it would be part of the NordVPN app when released. The broader implication is that security products will need to protect not only data and devices but also the autonomous software actors that increasingly mediate a user's digital life.

Small models, local processing and AI at NordVPN

Like most technology companies, NordVPN has built bespoke AI into its products while also using frontier models from major labs internally. For its next-gen antivirus feature, the company said it develops small models in-house to tackle specific tasks, such as spotting phishing attempts and other scams. Briedis said there are about 20 parameters that can be tweaked on the go, and the models work well right now. They can also run locally on a user's device. "You will not need a lot of hardware to do that these days, because they are really small," he said.

That local approach has obvious privacy and performance advantages. If a scam-detection model can run on a phone or personal computer, less data needs to leave the device. Local processing can also reduce latency and keep core protections working even when connectivity is poor. For a security company, the appeal is straightforward: smaller models can be purpose-built for narrow tasks, updated quickly and deployed across a wide range of consumer hardware.

Internally, NordVPN uses a blend of AI models but now often favors open-weight models. Like many companies, it started with frontier models from major AI labs, but costs quickly built up. With open-weight model performance improving, Briedis said NordVPN is not losing capabilities by switching. The result has been a surge in code production. However, that code goes through a thorough review process, which can create an approval bottleneck. "We are not doing the simple software development life cycle," he said. "We do secure software development lifecycle."

There are also limits. Briedis said one of NordVPN's non-negotiables is that teams handling payments do not use AI at all. Experimentation and innovation are reserved for other parts of the business. The caution reflects a broader tension in software development: AI can accelerate coding and reduce repetitive work, but it also introduces new review requirements, security risks and questions about accountability. Briedis noted that programmers across the industry are under pressure to adapt as their careers change shape. "I think every software development company right now is kind of changing how they work, and a lot of programmers [are] trying to figure out what they're going to do in five to 10 years right now, because their profession is changing so much," he said.

Privacy, trust and regulation

Given AI's growing unpopularity and its proven privacy risks, some security-conscious customers might feel uneasy about AI and privacy tools sharing space in a product. Briedis said NordVPN has not yet experienced customer pushback about AI in features like next-gen antivirus, partly because AI use is already common. "I cannot imagine people these days not using AI at all," he said.

He also stressed that the main way to sustain trust is not using customer data for model training, which he said NordVPN stands by. Protections such as the European Union's General Data Protection Regulation can add a layer of assurance on top of private company promises. On that front, however, Briedis thinks over-regulation, which he believes has happened in Europe, risks stifling innovation. He pointed to the Cyber Resilience Act as an example of compliance requirements that could make software development harder. "For example, the CRA Act that we're going to have in Europe, it's getting really hard for us to create software if we want to tick all those boxes to develop by that act," he said.

The dream tool: a personal security layer for digital exposure

When asked about his dream security tool for an AI-embedded future, Briedis described something broader than a VPN or an antivirus app. "For now, I think it's all about the information that is happening around you in your environment," he said. He envisioned an app or device that follows a user and tracks information connected to their name and what is happening on the internet around them as a person, making sure that data is interconnected, safe and secure. In effect, it would surveil every point of exposure, but for protection rather than exploitation. Ideally, such a product would safeguard a user's information and automatically delete it from the internet as required.

He acknowledged the limits of that vision. "The ideal scenario, I would say, is everything is on premises and on your machine, but that's not going to happen," Briedis said. Still, he outlined three pillars that would define a best-case scenario: data encrypted at rest, encrypted in transfer and, ideally, encrypted in use. The last is especially difficult without specialized hardware. Consumer devices such as phones and PCs are adapting to accommodate on-device AI features, but Briedis said NordVPN does not plan to branch into hardware to manage the attack surface anytime soon. "It was never our strategy to go into hardware, and I don't think that we're going to do it in the next couple of years, that's for sure," he said. "But you never know, you know? Maybe there's going to be a new, really affordable chip that you can use to create a device."

For a more immediate outlook on AI and security, Briedis believes the key is going smaller and staying local. "I think in the end, tech-savvy people are gonna go to their own LLMs, on their own computers," he said. "It's just a matter of time in my mind."

Source: ZDNET News