TL;DR
AI cannot build a pandemic, not because it is too weak, but because humans do not yet understand viruses well enough to teach it.
A team of scientists used AI to design several viruses that kill bacteria, then synthesised them in a lab. A 'genome-language model' is an AI trained on DNA sequences: feed it genomes, it hands genomes back. The loudest reaction to the news was fear. If it can make bacteria-killing viruses, can it make people-killing ones?
Last month Anthropic put out a report saying it had banned accounts using its Claude model for suspicious biological work. An AI that designs deadly viruses sounds frightening enough.
Why virologists are not spooked
The people who study viruses pushed back: this is not easy. What AI can design so far are only viruses that attack bacteria. Those are far simpler, and far smaller, than the viruses that attack humans.
To make a serious pathogen, you would first have to feed the AI data on what makes a virus more virulent and easier to spread. But virologists have not answered those questions themselves. Seema Lakdawala, a virologist at Emory, put it bluntly: right now the data going in is junk.
Even if the AI thinks it up, someone has to build it
Handed a pile of candidate sequences, a person still faces the hard labour: synthesise them, test them, confirm they are potent enough, produce a sufficient dose, and release the payload. That takes serious lab skill, expensive equipment and laboratory animals.
And viruses are fickle: give two experienced scientists the same protocol and the results may still differ. Try to dial up virulence and many hypotheses simply fail, sometimes destroying the pathogen entirely.
The thing actually worth worrying about is nature. Viruses keep revising themselves, unbound by human imagination, and need no villain to start an outbreak. Brandon Ogbunu, an evolutionary biologist at Yale, said he cannot look anyone in the eye and claim the AI process is more dangerous than what nature is already doing.
Still, he does not want AI regulated into uselessness. It can speed up vaccine development and viral surveillance. And virologists already run into blocked conversations when they ask Claude about flu genomes. 'It happens to me almost daily,' one researcher said.
In one line: AI is still just a tool. What we ask it, what we feed it, stays in human hands for now.
Curated from high-quality sources, with concise summaries and key takeaways.
