Article By Frank Bergman
Scientists have crossed a chilling new threshold after successfully using artificial intelligence (AI) to design and create entirely new, fully functional viruses that have never existed in nature.
Researchers from Stanford University and the Arc Institute used a generative AI system to create novel viral genomes from scratch before assembling them in a laboratory.
The result was 16 completely new viruses capable of infecting, replicating inside, and killing living bacterial cells.
Researchers insist the experimental viruses cannot infect humans.
However, biosecurity experts are sounding the alarm over what the breakthrough proves is now possible.
Machines can design functioning viruses.
And the safeguards needed to prevent the technology from being weaponized are not ready.
“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not,” warned Dr. Thomas Inglesby and Dr. Moritz Hanke of the Johns Hopkins Center for Health Security.
AI Generated 700,000 New Virus Designs
The bombshell research, now peer-reviewed and published in Science, used an AI model called Evo.
Rather than being trained on conventional human language, Evo was trained on billions of genetic sequences.
The system effectively processes DNA much like a large language model processes words, predicting which genetic building blocks should come next in a sequence.
Researchers focused on Phi X-174, a well-studied bacteriophage that infects E. coli.
They instructed Evo to generate new viruses using the genetic architecture it had learned.
The AI produced approximately 700,000 potential viral genomes.
Researchers narrowed those candidates to 300 and synthesized them in the laboratory.
Sixteen worked.
Those AI-designed viruses successfully infected bacteria, reproduced, and destroyed their hosts.
Synthetic biologist Patrick Cai of the University of Manchester described the achievement as “an important milestone.”
Scientists Have Crossed a Major Threshold
The medical possibilities are significant.
Bacteriophages can destroy bacteria, meaning AI-designed versions could potentially be developed to attack infections that have become resistant to antibiotics.
But the same breakthrough has an obvious darker side.
The technology has now demonstrated that generative AI can produce complete, viable viral genomes rather than merely analyze existing biological material.
In an accompanying Science commentary, Inglesby and Hanke warned that the question is no longer whether generative viral design will become possible.
It already is.
The issue now is whether humanity can control it before the same capability is directed toward pathogens capable of causing disease.
New disease-causing viruses “should not be pursued,” they warned.
Biosecurity Experts Sound the Alarm
Tom Ellis, professor of synthetic genome engineering at Imperial College London, noted that engineering complex human pathogens remains substantially harder than producing simple bacteriophages.
Ellis also argued that modifying existing pathogens would currently present a more realistic biological threat than designing an entirely new one through AI.
Nevertheless, experts warn that the direction of travel is unmistakable.
AI models are rapidly improving.
Genomic datasets are expanding.
Laboratory automation is becoming more sophisticated.
And the machines themselves can already generate viable biological organisms at a scale impossible for human researchers to match manually.
The experiment produced hundreds of thousands of candidate genomes in a single project.
Safeguard Depends on What AI Is Allowed to Learn
Researchers attempted to reduce the danger by restricting the biological information included in Evo’s training data.
But that safeguard is not inherent to the technology.
A different research group with access to broader datasets could train a similar model without the same restrictions.
That reality has alarmed biosecurity experts because controlling the information available to one model does nothing to eliminate the underlying capability.
Once researchers demonstrate that AI can generate functioning viral genomes, the knowledge itself cannot be erased.
AI Revolution Is Moving Into Biology
The development comes amid growing concerns about increasingly autonomous AI systems behaving in unexpected ways.
Major AI companies have recently revealed that advanced models were capable of exploiting cybersecurity vulnerabilities during controlled testing, including gaining access to external computer systems when safeguards failed.
Now the same accelerating technology has crossed into synthetic biology.
Instead of merely generating text, images, or computer code, AI can now write genetic code that becomes a functioning biological entity.
That represents an entirely different category of risk.
Population-Control Debate Adds Darker Context
The breakthrough also comes as scientists and policymakers continue pushing proposals to deliberately drive down global population levels in the name of environmental “sustainability.”
A recently published study called for gradually reducing humanity’s population to approximately four billion people by 2200.
That proposal was framed around voluntary reductions in fertility and consumption.
There is no evidence connecting those researchers to the new virus-design work.
Nevertheless, the collision of these two developments highlights the enormous stakes surrounding technologies capable of engineering biology.
A system powerful enough to create beneficial viruses is also a system whose misuse could have catastrophic consequences.
The danger becomes considerably more serious when biological engineering technology is combined with political ideologies that regard dramatic population reduction as desirable.
‘Governance to Safely Steer It Does Not’
The Stanford and Arc Institute researchers demonstrated something that until recently belonged largely to science fiction.
Artificial intelligence can design a virus.
Scientists can take that machine-generated genetic blueprint and construct it.
And the resulting organism can live, reproduce, infect its target, and kill it.
For now, those targets are bacteria.
But biosecurity experts are warning against assuming that boundary will remain permanent.
“The ability to compose viral genomes using generative AI now exists,” Inglesby and Hanke warned.
“The governance to safely steer it does not.”
Humanity has entered an era in which machines can write biological code capable of becoming life-like, self-replicating infectious agents.
The technology may ultimately produce powerful new medicines.
But the same breakthrough means the tools needed to design something far more dangerous are moving rapidly closer to reality.
And unlike the technology, the safeguards are already falling behind.
Be the first to comment