AI Just Designed Viruses That Did Not Exist in Nature. What Happens Next?

Artificial intelligence has crossed another remarkable and uncomfortable frontier.

Researchers have used generative AI to design entirely new viruses, and some of those designs successfully functioned in laboratory experiments.

The development is being described as a major milestone for generative biology. It also raises a difficult question for the technology industry: when AI becomes capable of designing biological systems, how do we make sure the same capability cannot eventually be turned toward harm?

Researchers from Stanford University and the Arc Institute used genomic AI models from the Evo family to design new bacteriophages—viruses that infect bacteria rather than humans. In laboratory testing, 16 of 285 tested AI-generated designs successfully propagated and inhibited the growth of the targeted bacterial https://arcinstitute.org/news/evo-2-one-year-laterstrains

The research focused on bacteriophages, and the researchers deliberately incorporated safety measures into the work. The resulting phages were designed to target E. coli, and the team reported restricted host specificity in its experiments. Arc also says that eukaryotic viruses were excluded from Evo 2’s training for safety reasons.

So why is everyone paying attention?

Because this isn’t simply a story about another AI tool. It is a story about AI learning to work with the language of life.

From predicting biology to designing it

Traditional AI applications have largely operated inside the digital world.

They write text. Generate images. Analyse data. Write software.

Now, increasingly capable biological foundation models are being used to analyse and generate genetic sequences.

Evo 2, developed by researchers at the Arc Institute in collaboration with Stanford, UC Berkeley, UCSF and NVIDIA, was designed to understand biological sequences at a scale that allows it to make predictions and generate genetic code. Arc describes it as a model capable of working across genomes from different domains of life.

That changes the conversation.

The question is no longer simply: “What can AI tell us about biology?” It is becoming:

“What can AI help us create in biology?”

And that is where the story becomes considerably more complicated.

The breakthrough is also the warning

The researchers’ achievement demonstrates something important: generative AI can contribute to the design of complete biological systems rather than merely analysing isolated pieces of genetic information. That has enormous potential.

AI-designed bacteriophages could eventually contribute to research into phage therapy, particularly as antibiotic resistance continues to make some bacterial infections increasingly difficult to treat. Arc says several of its experimentally validated designs were capable of inhibiting the growth of targeted bacterial strains, illustrating the potential of AI-assisted biological design.

In other words, the same capability making people nervous could also help researchers solve serious medical problems. That is the paradox.

The technology is not inherently good or bad.

Its significance comes from what humans choose to do with it.

But what happens when the technology becomes more capable? This is where the biosecurity conversation becomes unavoidable.

The current experiment does not demonstrate that AI can create a virus capable of infecting humans. It does not show that an AI system has independently produced a human pathogen. And it certainly does not mean that an AI-generated pandemic is around the corner.

But it does demonstrate something that would have sounded considerably more futuristic only a few years ago: AI can now participate in designing biological systems that can be experimentally validated.

As biological AI models become more capable, researchers, policymakers and technology companies will need to think about safeguards alongside capability not after capability has already advanced.

Arc itself acknowledges that future versions of biological foundation models will require continued attention to biosafety as their capabilities develop.

AI security can no longer be understood solely through the lens of passwords, data breaches, malicious code or phishing. As AI becomes integrated into scientific research, engineering, healthcare and biotechnology, the consequences of AI misuse can extend beyond the digital environment.

That means security has to evolve alongside capability. A model that can generate text requires one type of safeguard. A model that can generate software requires another. A model capable of designing biological sequences introduces an entirely different category of risk.The more powerful the system becomes, the more important it becomes to ask not only:

“Can AI do this?” but also: “Should it?”

And perhaps most importantly:“What safeguards exist if someone tries to misuse it?”

Innovation needs security built into it

The AI-generated virus story is therefore neither simply a breakthrough nor simply a threat.

It is both a scientific achievement and a security conversation.

The potential benefits are significant: faster biological discovery, new approaches to antibiotic-resistant infections, better understanding of genetic systems and eventually new possibilities in medicine and biotechnology.

But the risks cannot be treated as an afterthought.

As AI moves from generating information to influencing the physical world, security must move with it.

The lesson is bigger than viruses. AI is becoming increasingly capable of creating things that never existed before.

The next challenge is ensuring that our ability to secure those creations develops just as quickly as our ability to generate them.

The ESET Perspective

For organisations adopting AI, this development is another reminder that AI governance and cybersecurity can no longer be treated as separate conversations.

The systems businesses introduce today may eventually influence decisions, code, research, infrastructure and other real-world processes.

That makes questions around access, oversight, data protection, model security, responsible use and human accountability increasingly important.

AI’s capabilities are accelerating.

Security needs to keep pace.

What Does This Mean for Cybersecurity?

The significance of AI-designed biological systems extends beyond the laboratory.

It is another reminder that artificial intelligence is rapidly moving from a tool that analyses information to a technology capable of creating, adapting and influencing increasingly complex systems.

For cybersecurity, that evolution is already visible.

Cybercriminals are using AI to make attacks faster, more convincing and more scalable. ESET’s H1 2026 Threat Report, for example, reports the growing use of AI in malicious activity, including AI-themed social engineering, malicious AI skills and emerging AI-assisted malware.

That creates a race of its own.

As AI gives defenders new capabilities, it can also give attackers new ones.

This is where ESET’s approach becomes particularly relevant. ESET has been applying machine learning to cybersecurity for decades, combining machine-learning models with technologies such as behavioural analysis, cloud reputation systems and multilayered detection rather than relying on a single security mechanism.

More recently, ESET has expanded its AI work beyond traditional threat detection. Its ESET AI Advisor uses proprietary generative AI alongside security telemetry to help security analysts investigate and respond to threats more efficiently, while ESET is investing further in cybersecurity-specific AI and AI security capabilities.

The connection to this latest biological AI breakthrough is therefore not that ESET is protecting people from AI-generated viruses. It isn’t.

The connection is bigger:

AI capabilities are accelerating across industries. Security has to accelerate with them.

Whether AI is being used to analyse a cyber threat, generate software, design biological sequences or automate an entirely new workflow, the fundamental question remains the same:Can we innovate at the speed of AI without leaving security behind?

That is the conversation this breakthrough should start.Because the future of AI security will not be defined only by what artificial intelligence can create.

It will also be defined by how responsibly we learn to secure what it creates.

And that’s where cybersecurity companies like ESET have an increasingly important role to play.