AI Can Evade DNA Biosecurity Screening

AI Circumvents Biosecurity Measures

A recent study reveals that artificial intelligence can create DNA for various hazardous proteins in a way that evades detection by biosecurity screening methods used by DNA manufacturers.

Major biotech firms that produce custom DNA for scientists have safeguards to prevent dangerous biological materials from falling into the wrong hands, screening orders to identify those seeking genes for substances like smallpox or anthrax. However, research published in the journal Science shows how AI can bypass these biosafety protocols.

How AI Rewrites DNA

AI researchers discovered that protein-design tools could "paraphrase" the DNA codes of toxic proteins, rewriting them while maintaining their structure and potential function, according to Eric Horvitz, Microsoft's chief scientific officer. Using an AI program, computer scientists generated DNA codes for over 75,000 variants of dangerous proteins, and the firewalls used by DNA manufacturers often failed to detect them.

Horvitz noted that these reformulated sequences bypassed the biosecurity screening systems used globally by DNA synthesis companies to identify dangerous orders. Although a quick fix was implemented in the biosecurity screening software, it was not entirely effective, still missing a small percentage of the variants.

This incident highlights growing concerns about the potential misuse of powerful biological tools enhanced by AI. Horvitz stated that AI-powered protein design is a promising field with advancements in medicine and public health, but like many powerful technologies, it can be misused.

Biosecurity Concerns

Biologists have long been concerned that increasingly sophisticated DNA tools could be used to create potent biothreats, such as more virulent viruses or easily spread toxins. There have been debates about the wisdom of openly publishing certain experimental results, despite the importance of open discussion and independent replication in science.

The researchers and the journal publishing this study decided to withhold some information and restrict access to their data and software, enlisting the International Biosecurity and Biosafety Initiative for Science, a non-profit organization, to manage access based on legitimate need. Horvitz noted that this is the first instance of such a model being used to manage the risks associated with sharing hazardous information in a scientific publication.

Arturo Casadevall, a microbiologist and immunologist at Johns Hopkins University, praised the work, noting that it identifies and attempts to correct known vulnerabilities. However, he also raised concerns about unknown vulnerabilities that may require future corrections. Casadevall also pointed out that the team did not conduct lab work to confirm whether the AI-designed proteins would mimic the activity of the original biological threats. While such work would provide a crucial reality check, it might be prohibited by international treaties against biological weapons development.

Past Research

Previously, scientists have explored the potential for AI to be used for malevolent purposes in biology. For instance, a team investigated whether AI could generate novel molecules with the same properties as nerve agents. In less than six hours, the AI tool created 40,000 molecules that met the criteria, including known chemical warfare agents like VX and many unknown molecules predicted to be more toxic.

The researchers wrote that they had transformed their innocuous generative model from a helpful tool of medicine to a generator of likely deadly molecules. They refrained from openly publishing the chemical structures or creating them in the lab due to the perceived danger, according to David Relman, a researcher at Stanford University. Relman views the current study as a warning and commends its efforts to address security screening vulnerabilities. However, he also emphasizes the magnitude of the problem and the need to proactively address it.

Reassurance from Experts

Despite these concerns, some biosecurity experts express reassurance. James Diggans, head of policy and biosecurity at Twist Bioscience and chair of the board of directors at the International Gene Synthesis Consortium, stated that Twist Bioscience, a major provider of custom DNA, has had to refer orders to law enforcement fewer than five times in the past decade.

Diggans emphasized the rarity of misuse attempts in biotech, contrasting it with the cybersecurity world. He suggests that the number of individuals attempting to misuse biotech tools may be very low, making these security systems an important safeguard, and indicating that such scenarios are uncommon.