AI-Generated Bioweapons: An Imminent Danger. Urgent Government Action Needed.

Published Date: 14/09/2024

Powerful new AI models, combined with novel lab tools, could make it easier to create deadly viruses. Governments, tech companies, and scientists must act to lower the risk.

As deadly as the COVID-19 pandemic was, the next one could be more nightmarish. Powerful new artificial intelligence (AI) models, combined with novel lab tools, could soon enable rogue scientists or states to engineer a pathogen that would spread faster, resist vaccines better, and kill more people than COVID-19 did.

  Nature has always had the ability to concoct nasty pathogens, from the plague to the Spanish Flu. For many decades, so have humans  The Japanese conducted brutal biological warfare experiments in World War II, and the US and the Soviet Union stockpiled toxins during the Cold War, with the latter’s program continuing even after signing the Biological Weapons Convention in 1972.

  However, such efforts have traditionally been limited by the number of scientists trained to conduct the necessary research and the tools available for producing and distributing effective weapons. Technology is breaking down both barriers. Large language models (LLMs) such as OpenAI Inc’s ChatGPT can synthesize vast amounts of knowledge rapidly  In one experiment, a chatbot advised a group of Massachusetts Institute of Technology students about how to engineer four potentially deadly pathogens and where to procure the necessary DNA without detection — in an hour.

  More specific AI programs trained on biological data, known as biological design tools, are even more powerful. Over time, such programs could speed the development of entirely new pathogens with deadly properties, perhaps even the ability to target specific populations. Emerging technologies — from “benchtop” synthesizers that would allow individual researchers to create their own strands of DNA, to so-called cloud labs where experiments can be conducted remotely using robots and automated instruments — would lower other hurdles to testing and producing potential weapons.

  It is worth noting that the barriers to producing and distributing a workable weapon remain quite high. However, scientists say that could change in a few years, given how fast all these technologies are progressing. The time to act is now, before they reach maturity.

  A series of interventions would help. Begin with the AI models. Some US developers of the most powerful LLMs are voluntarily submitting them to the government for further evaluation. That is welcome, but more scrutiny might be warranted for the riskiest models — those trained on sensitive biological data. Congress should work with AI developers and scientists to develop criteria for which models might require formal screening and what guardrails could be included in those found to pose the highest risks.

  The next task is to prevent any AI-designed viruses from entering the real world. Providers of synthetic nucleic acids should be required to know their customers and screen orders for suspicious DNA sequences. All requests should be logged, so new pathogens can be traced back if they are released into the wild. Controls should also be built into benchtop synthesizers, while cloud labs should scrutinize customers and requests. Risky experiments should always have a human in the loop.

  The US should press other countries to adopt similar safeguards, so rogue actors cannot simply seek out less scrupulous providers elsewhere. If the Biological Weapons Convention cannot be toughened, because of diplomatic frictions, like-minded countries should at least agree on a set of best practices, as they have begun to do with AI.

  Above all, countries ought to harden their pandemic defenses so that anyone who manages to exploit loopholes in the system cannot cause extensive damage. AI itself could boost the ability of governments to detect the emergence of new pathogens, not to mention speed the development of vaccines, and the production and distribution of personal protective equipment. Stronger public health systems are critical, whether new viruses are produced by terrorists, rogue states, accidents, or nature.

  COVID-19 exposed huge gaps in those defenses, too many of which remain unfilled. Governments have every incentive to head off this new threat while there is still time.

FAQS:

Q: What is the main concern about AI-generated bioweapons?

A: The main concern is that powerful new AI models, combined with novel lab tools, could soon enable rogue scientists or states to engineer a pathogen that would spread faster, resist vaccines better, and kill more people than COVID-19 did.

Q: What is the role of large language models (LLMs) in bioweapon development?

A: LLMs such as OpenAI Inc’s ChatGPT can synthesize vast amounts of knowledge rapidly, making it easier to engineer potentially deadly pathogens and procure the necessary DNA without detection.

Q: What are biological design tools, and how do they contribute to bioweapon development?

A: Biological design tools are AI programs trained on biological data that can speed the development of entirely new pathogens with deadly properties, perhaps even the ability to target specific populations.

Q: What measures can be taken to prevent AI-designed viruses from entering the real world?

A: Providers of synthetic nucleic acids should be required to know their customers and screen orders for suspicious DNA sequences, and controls should be built into benchtop synthesizers and cloud labs.

Q: How can governments harden their pandemic defenses against AI-generated bioweapons?

A: Governments can press other countries to adopt similar safeguards, agree on a set of best practices, and harden their pandemic defenses by strengthening public health systems and leveraging AI to detect the emergence of new pathogens and speed the development of vaccines and personal protective equipment.

More Topics :