Researchers use AI to design functional viruses and warn of misuse risks

The experiment relied on a genetic library that allowed the model to «learn the evolutionary constraints» of natural genomes. However, the scientists discovered that one of the generated viruses exhibited an aspect «evolutionarily distant» from any known pattern, a sign that the AI may have produced changes that natural evolution might not have revealed as possible for millions of years.

A method inspired by language models

The authors emphasize that these results demonstrate the growing ability of generative models to design viable biological agents from little initial information. They warn that the same method, theoretically accessible to malicious actors, could be misused to create dangerous pathogens, or even for bioterrorist purposes.

The study, published in a peer-reviewed scientific journal, does not specify the exact identity of the AI model used, but the methodology resembles that of large language models such as ChatGPT. The researchers advocate for safeguards, including controlled access to genetic databases and enhanced oversight of DNA synthesis experiments.

Asked about the implications of this work, geneticist Jordi García Ojalvo said it was becoming «urgent to establish clear safety protocols» before the technology spreads more widely. In his view, the scientific community must anticipate potential abuses rather than wait for an incident to occur.

Calls for stronger oversight

The Science Media Centre, a British organization that relays scientific expertise to the media, said it was examining the study and expected to publish a detailed comment from independent experts shortly. Several voices within the biosecurity community are calling for a public debate on the regulation of AI applied to synthetic biology.

At the Johns Hopkins Center for Health Security, specialists recalled that scenarios involving the misuse of AI for biological purposes had already been raised in previous reports and that this threat should be treated seriously. The institution has for several years advocated international cooperation to prevent the creation of dangerous synthetic pathogens.

The announcement reignites the broader debate on the oversight of generative artificial intelligence, as American and European regulators discuss new rules. Figures from the technology sector, including Eric Schmidt, had already alerted public opinion to the potential risks associated with models capable of designing biological agents, underscoring the need for governance suited to these new capabilities.

A broader debate on AI regulation

The researchers from Stanford and the Arc Institute say that transparency about their methods is a first step toward a collective risk assessment. They announce the upcoming publication of a framework of best practices intended for laboratories working on AI applied to synthetic biology.

According to the authors, the main lesson of the study is that AI models are now capable of producing functional biological entities from little information, which, in their view, requires a revision of existing biosafety protocols and close cooperation between researchers, public authorities, and industry.