Why AI Probably Won’t Annihilate Humanity Using Bioweapons

“Currently, it’s impossible for AI to independently operate a fully autonomous lab and create a virus, as such labs do not yet exist,” states Olivia Scharfman, a biotechnology fellow at the Institute for Progress. “However, I believe an AI could hire someone to accomplish that task.”

She views AI-assisted bioterrorism as an urgent concern, believing that some groups, including “transhumanist AI successionists”—adherents of a fringe ideology aiming to replace humans with machines—might pursue such actions.

Conversely, other experts express less concern over AI-related bioweapons, arguing that biological warfare lacks methodological sensibility. “The associated risks are often misunderstood,” comments genetic biologist and computational biology professor Francois Belloux to WIRED, suggesting that people tend to “overrate the effectiveness of pathogens as weapons.”

Setting aside the issue of AGI, Belloux emphasizes that bioweapons are not particularly appealing options for those aiming for mass harm, as targeting specific groups while avoiding others is challenging, and the logistics of producing and distributing pathogens are far more complicated than, for instance, deploying a bomb. “If your goal is to kill, there are far more effective methods than engineering a virus or bacterium and releasing it,” he states—regardless of whether one is a highly advanced AI or an ordinary human.

With AI highlighting the potential risks of bioweapons, Scharfman believes now is the time to advocate for enhanced protections against all biological threats, including existing pathogens like H1N1. This could encompass significant measures such as enacting legislation to tighten security protocols around DNA synthesis technologies, as well as upgrading air purification systems in buildings to reduce risks from all viruses.

Determining whether AI genuinely increases the risk of bioweapon creation is complex, according to Steph Guerra, head of AI and bio at the Rand Corporation. “AI excels at integrating information and motivating individuals towards both beneficial and harmful actions,” she explains.

Even with a low risk, Guerra points out that society can still take steps to guard against possible misuse. One approach is for governments to pass regulations requiring companies that sell synthetic DNA and RNA to vet customers and orders, helping to prevent genetic material misuse. Many companies currently screen orders for “sequences of concern” using software, though this practice is not universally adopted or mandated.

Before any malicious actor could reach that stage, however, there’s a consensus that models must include adequate safeguards to prevent them from providing harmful or actionable information.

Guerra also suggests that enhancing global surveillance systems to quickly detect new disease outbreaks can reduce the risk of an AI-generated biothreat, emphasizing the importance of data-sharing protocols among AI companies, gene synthesis providers, and government entities to mitigate misuse.

“With almost any biosecurity measure, there will always be ways to bypass them,” she states. “This is why we require layered strategies that create friction throughout the entire process, from ideation to the intent of a malicious actor, all the way to the release of a bioweapon.”

Unutmaz raises another critical concern: the prevalent discussions around AI dangers might distract from the potential benefits these advanced systems could offer in vaccine development and other medical innovations.

“We really need to concentrate on its positive aspects,” he emphasizes.