The possibility of artificial intelligence being used to create dangerous biological agents is becoming a growing concern among scientists, as advances in AI allow researchers to design new viruses in ways that were previously impossible.
A study published in Nature has revealed that scientists successfully used an AI model called Evo to create 16 new viruses in a laboratory setting. The viruses were bacteriophages, meaning they were designed to infect bacteria, specifically E. coli, and do not pose a direct threat to humans.
What has raised concerns, however, is the technology behind the achievement. Researchers trained AI to interpret genetic sequences and use that knowledge to generate entirely new viruses capable of self-replication, as reported by Stuff.
University of Otago Professor Nick Wilson, who has studied existential risks for more than three decades, said safeguards were urgently needed.
“We urgently need guard rails put in so that this type of research can only be used for good and that it's done in highly supervised situations,” he said, Stuff has quoted.
According to a reprot by Stuff, AI-powered biology could have major benefits for medicine. Bacteriophages, for example, can potentially be used to deliver treatments for genetic diseases, while increasingly advanced AI systems are already helping scientists tackle difficult problems in biology and mathematics.
However, experts are also concerned that the same technology could eventually be used to develop pathogens capable of causing widespread harm.
“The Justinian plague in about 500 AD possibly killed 50% of the European population. So that's what a natural pathogen can do. A souped-up pathogen could theoretically kill much more than 50%,” Wilson said, Stuff has quoted.
The threat is not limited to deliberate biological warfare. Researchers are also concerned about accidents, mistakes or pathogens behaving in unexpected ways after being developed or released.
“Say a country like North Korea produced a pathogen that was designed to wipe out all the cattle in America and would be prepared for its tiny cattle herd to be wiped out as well - but then that particular pathogen jumps from cattle to humans,” Wilson said, as quoted by Stuff.
History has shown that laboratory accidents involving dangerous pathogens can have serious consequences. In 1978, medical photographer Janet Parker became the last known person to die from smallpox after exposure to a laboratory sample at a university in Birmingham.
AI companies are increasingly testing their systems for potential misuse in biological and chemical weapons development. However, the rapid progress of the technology is creating new challenges for safety researchers.
According to a report by Stuff, Anthropic, the company behind Claude, has publicly released research examining the capabilities of its latest model, Mythos. Researchers found that the system could significantly enhance the ability of people with existing technical knowledge to work on biological or chemical weapons.
“It has the ability to significantly help individuals or groups with basic technical backgrounds (eg undergraduate STEM degrees) create/obtain and deploy chemical and/or biological weapons with serious potential for catastrophic damages,” researchers said, Stuff has quoted.
They also cautioned that AI does not replace most forms of advanced human expertise, but could speed up work by highly skilled teams.
During testing, teams of generalist microbiologists using Mythos completed a laboratory task involving a novel plant pathogen much faster than specialist teams working without AI assistance. Researchers estimated the task could have taken up to three months without AI, while two-person teams using Mythos completed it in 16 hours, as reported by Stuff.
AI companies have safeguards designed to prevent ordinary users from accessing such capabilities. However, concerns remain about how quickly AI systems are advancing and how difficult increasingly powerful models may be to predict.
Experts say voluntary safeguards by technology companies may not be enough as major AI developers in the US and China compete to build increasingly powerful systems.
“The divergence between the technology and the controls just keeps getting worse. The gap is just enormous,” Wilson said, as quoted by Stuff.
He also warned that existing international frameworks have failed to keep pace with developments in AI and biotechnology.
“There's no really tightly agreed controls on AI and the international biological weapons conventions are hopelessly out of date and inadequate and haven't got strong enforcement mechanisms,” Stuff has quoted.
One proposed response is to strengthen the Biological Weapons Convention, a treaty that has prohibited the development, production and stockpiling of biological weapons since 1975.
Countries including the US and China are considering how the treaty could address the emerging risks posed by AI-assisted biological warfare. However, practical measures for enforcing new rules remain under discussion.
According to Stuff, another possible safeguard is tighter screening of orders for genetic material that could potentially be used to create dangerous pathogens. Some suppliers already screen genetic sequences, although research has suggested AI could potentially be used to disguise problematic orders.
One of the emerging strategies for dealing with the threat is to use AI itself as a defence mechanism. Specialised systems could be developed to detect biological threats more quickly and help scientists develop vaccines and treatments in response, Stuff has reported.
Using AI to defend against AI-designed pathogens may sound like science fiction, but experts argue that rapidly advancing technology could make such measures increasingly necessary.
As AI capabilities continue to expand, scientists and policymakers face a race of their own, ensuring that safeguards and international rules develop quickly enough to keep pace with the technology, as reported by Stuff.