Australian scientists assess AI risks for biological weapons — ABC News Australia
In Australia, researchers in biosafety and public health warn that the development of artificial intelligence systems could lower barriers to creating or using dangerous biological agents. At the same time, these technologies are already being used in medical and agricultural research, ABC News Australia reports.
Anthropic CEO Dario Amodei previously called for slowing AI development so that technology does not outpace humanity’s ability to control it. One of the cited risks is the potential use of models to develop dangerous biological agents.
Lowering barriers to dangerous research
Southern Cross University security researcher Brendan Walker-Munro believes that AI lowers barriers to accessing specialized knowledge. University of New South Wales public health scientist David Heslop said that the speed at which such tools are developing complicates human oversight.
According to him, large language models can help malicious users plan the use of dangerous materials, including already existing viruses or toxins. At the same time, Heslop noted that this still requires professional training and laboratory equipment costing tens of thousands of dollars.
In 2022, researchers at Collaborations Pharmaceuticals demonstrated that AI software designed to screen out potentially toxic drugs could be repurposed to design dangerous toxins. Anthropic also reported in its threat intelligence report five cases of its products being used in ways that could support the development of biological weapons. The company blocked those users for violating rules concerning supported regions.
More current news is available on the UA.News Telegram channel Telegram.
Benefits for medicine and agriculture
AI is also being used for beneficial biological research. The AlphaFold program has been used to predict protein structures in medical development; its creators received the 2024 Nobel Prize in Chemistry. US scientists have also used generative AI to create new viruses that infect bacteria and could potentially be used against infections.
University of Queensland computational biologist David Kainer and his colleagues developed an autonomous AI tool for agricultural research. It processes large datasets and models how environmental or genetic changes can affect plant traits, including their yield and drought resistance.
Safeguards and oversight
Commercial language models have restrictions intended to prevent their use for dangerous purposes. Heslop believes that laws and regulation are needed alongside such safeguards. Walker-Munro stressed that open-weight models can be downloaded and run privately by people with basic programming skills.
Among possible measures, experts cite tighter control over materials and equipment, as well as the use of AI to oversee sensitive research. At the same time, Walker-Munro warned that excessive restrictions could hinder legitimate scientific work.
