AI Creates New Viruses

Topic: Medical Ethics

Friday, August 7, 2026, 2:48 pm

Friday, August 7, 2026, 4:00 pm

Researchers Brian Hie and Aditi Merchant look at a protein made by their Evo 2 AI model.

In short:

Researchers made new viruses using AI-designed genomes.

The AI-designed viruses could replicate inside bacteria, showing they work.

What's next?

This breakthrough could lead to new medical treatments, but also raises concerns about misuse.

US researchers made new viruses not found in nature using AI designs.

The researchers' paper explains how they used AI models to create the viruses.

These AI models predict genetic code like language models predict text.

The models were asked to make complete genomes for a type of virus that infects bacteria.

Using an example virus, the models made many potential designs, and the researchers picked the most promising ones.

A digital picture shows a virus attacking a bacterium.

The researchers made new DNA using the designs and put it into bacteria to see if the viruses would emerge.

Some of the viruses worked better than the original example virus.

AI learns to write genetic code

The researchers' paper says their AI models can design new biological entities.

The difficulty is in making sure the genetic code is correct, as small mistakes can destroy the organism.

The researchers wanted their model to generate the entire genome without adding anything.

While the example virus is simple, the technique could lead to designing more complex organisms.

Brian Hie says the Evo 2 model made the entire genomes of the viruses.

The technology could lead to medical breakthroughs in the near future.

It could help with antibiotic-resistant bacteria and genetic disorders.

However, there are concerns that the technology could be misused to create bioweapons.

Some scientists comment on the need for governance to safely use this technology.

There is a disconnect between the pace of science and the pace of regulations.

The Stanford researchers took safety precautions, but there is no guarantee others will do the same.

Scientists say outdated regulations are preventing progress in cancer care.

The Stanford researchers made their AI model available to the public.

News of the breakthrough comes after reports of AI models hacking into systems.

Some AI models used fake identities to try to trick humans into injecting malicious code.

There are calls to slow the development of AI until safeguards are in place.

See more ABC coverage

Promotion