Top AI executives are sounding the alarm, pushing Congress to create new laws that would make it tougher for malicious actors to weaponize AI for biological threats. CEOs from OpenAI, Google DeepMind, Anthropic, and Microsoft AI are among those who signed a public letter advocating for stricter screening of customers and orders for synthetic DNA and RNA.
The letter, spearheaded by the Institute for Progress and the Foundation for American Innovation, acknowledges the rapid advancements in AI could significantly lower the barriers to creating biological weapons. Historically, the complex knowledge and resources required have been a deterrent, but that might change soon.
The science of synthesizing DNA, first achieved in the 1950s, is now largely automated. Numerous companies worldwide offer custom genetic sequences for research, drug development, and diagnostics. While many providers already vet their clients, not all consistently screen customers or the specific genetic material they order, raising concerns about potential misuse.
This isn't just theoretical. In 2017, researchers successfully reconstructed an extinct horsepox virus using mail-order DNA, a feat that critics warned could be replicated with deadlier viruses like smallpox. Since then, gene synthesis has become even more affordable, and coupled with AI's capabilities, it's becoming increasingly feasible to design novel toxins and pathogens, although creating a functional virus from scratch would likely still require some biological expertise.
David Relman, a biosecurity expert from Stanford University who signed the letter, highlighted how AI tools could help users find screening-free sequences or even modify orders to evade detection. The signatories, a mix of scientists, national security experts, and executives from gene synthesis companies like Twist Bioscience, are advocating for robust regulations. Many in the industry already support voluntary screening for dangerous gene sequences.
The push for formal rules comes as existing measures face scrutiny. Federal guidelines require federally funded researchers to use screened providers, and a bipartisan Senate bill aims to mandate screening for all U.S. gene synthesis providers. However, a recent Microsoft study revealed that AI protein design tools could generate dangerous sequences that bypassed current screening software. Experts like Geoff Ralston, former president of Y Combinator, argue that AI labs developing biological models should implement their own user screening, making it extremely difficult to misuse the technology for harmful purposes.
