New Study Uncovers AI’s Potential to Misuse DNA Technology
Major biotech firms, tasked with producing custom DNA for researchers, have safeguards to prevent dangerous biological materials from falling into the wrong hands. They actively monitor orders to stop anyone looking to acquire harmful agents like smallpox or anthrax genes.
However, a recent study published in the journal Science highlights how artificial intelligence could easily bypass these safety measures.
Researchers discovered that AI-driven protein-design tools can essentially “rewrite” the DNA sequences of harmful proteins. By altering the codes while still preserving their structure—and possibly even their function—this technique poses new challenges for biosecurity. Eric Horvitz, Microsoft’s chief scientific officer, notes this capability with concern.
The scientists employed AI to generate over 75,000 DNA variants linked to dangerous proteins. Notably, DNA manufacturers’ firewalls failed to detect a significant number of these variants. “To our concern,” Horvitz remarks, “these reformulated sequences slipped past the biosecurity screening systems used worldwide by DNA synthesis companies.”
A quick fix was implemented in the biosecurity software, yet it still couldn’t catch a small percentage of those variants. It’s, well, a temporary solution to an ongoing issue about the potential for misuse of advanced biological tools fueled by AI.
The Risks Associated with Open Science
AI-driven protein design is a promising area in scientific research. As Horvitz points out, there are remarkable advancements in medicine and public health. However, the flip side is that these powerful tools can also be misused.
For years, biologists have expressed concerns regarding the potential to engineer dangerous biological threats, such as more infectious viruses or easily transmissible toxins. There has been debate over whether it was wise to openly share certain experimental findings, as open discussion is vital for scientific advancement.
This recent study led researchers to be cautious about the information they shared, opting to restrict access to their data and software through a third party. Horvitz describes this as the first instance of such a model being utilized to manage risks surrounding the dissemination of sensitive information in scientific publications.
Response from the scientific community has been generally positive. Arturo Casadevall, a microbiologist at Johns Hopkins University, expressed a favorable reaction, stating that the study identifies vulnerabilities and shows an effort to rectify them. Yet, he raises a valid concern: “What vulnerabilities don’t we know about that will require future corrections?”
Interestingly, the study did not include lab experiments to determine whether the AI-generated proteins would truly mimic the actions of the real threats. A reality check like this could provide insights into how society will handle emerging AI-related biosecurity issues—though, of course, executing such experiments could conflict with international agreements against creating biological weapons.
Staying Ahead of Rapid AI Development
This isn’t the first instance researchers have pondered the potential misuse of AI in a biological context. A few years back, another team explored if AI could generate novel molecules mirroring nerve agents. In a remarkably short timeframe, the AI crafted 40,000 molecules fitting the criteria, including both known and new chemical warfare agents.
Concerns arose regarding the associated risks, prompting that team to refrain from publishing the chemical structures their AI drafted or creating them in the lab, as they felt it would be too perilous.
David Relman, a Stanford researcher, commends the latest study for showcasing how AI could evade security measures. Yet, he also warns of the considerable challenges that lie ahead. “How do we get ahead of a freight train that is just evermore accelerating?” he wonders.
Despite concerns, some biosecurity experts find reasons for optimism. James Diggans from Twist Bioscience, a leading custom DNA provider, mentions that orders needing referral to law enforcement have been rare—fewer than five times in the past decade. “The reality is that few people seem motivated to misuse biotechnological advancements,” he states, emphasizing that we should not overlook the effectiveness of existing systems.
