AI Intensifies Biological Weapon Risks: Tech Giants and Scientists Issue Warnings

Overview

Recently, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have publicly stated that the development of artificial intelligence could bring serious risks, advocating for careful control over the pace of cutting-edge technology advancement. At the same time, multiple researchers have warned that the risk of AI fostering biological weapons is imminent, with a rare consensus forming among tech giants and scientists.

AI Lowers the Barrier to Biological Weapons Research

Researchers are concerned that AI tools could be used to assist in the design, manufacture, and deployment of new biological agents. In 2022, a research team at Collaborations Pharmaceuticals discovered that using an AI molecular generator, it was possible to generate 40,000 potential chemical warfare agent molecules in just six hours, some of which were even more toxic than known nerve agents. This finding was once seen as a wake-up call in the field, but the situation is now even more dire.

• Biological weapons come in various forms, including highly lethal viruses targeting specific genes, fungi that destroy crops and trigger food crises, or undetectable colorless and odorless toxins.
• Large AI models incorporate the knowledge reserves of almost all scientists throughout history and can provide experimental guidance as well as video training.
• As gene editing and synthetic biology technologies become increasingly widespread, the "DIY bio" movement has made home laboratories feasible, lowering the technical threshold.

"Anyone with determination may eventually succeed." said Dunya Sabra, a biosafety researcher at the University of Hamburg, noting that as the technological threshold drops, the potential threat will rise exponentially.

Effectiveness of Existing Protective Measures Questioned

Currently, the industry has established certain protective mechanisms. DNA synthesis companies screen suspicious orders, and responsible researchers evaluate risks and develop mitigation measures through "red team" and "blue team" testing. AI companies have also adjusted their tools to try to prevent them from providing scientific information that could be misused. However, these protections are not foolproof.

Anthropic's report released last week acknowledged that individuals had attempted to use its models to explore methods such as enhancing the transmission capacity of the chikungunya virus, creating more dangerous mutant strains of avian influenza, and constructing "venom toxin peptide maps." Professor David Magnus of Stanford University stated that this is an ongoing game, because AI excels at finding ways to evade monitoring, and in the future, AI may need to be used to restrict the use of AI.

Although some scientists at Imperial College believe that current AI is not yet capable of independently completing the full-cycle development of biological weapons, and that human work in laboratories remains crucial, Sabra emphasized the need for forward-looking preparedness. She recommended that countries strengthen their healthcare systems and stockpile antidotes and medications for known toxins to address potential crises.