Artificial intelligence has moved from writing text and code to writing genomes.
According to Ars Technica, researchers have used large genome models — AI systems trained on genetic sequence data rather than human language — to design new viruses. The report describes a system that produced genetically distant versions of a bacteria-killing virus, meaning the AI's output was not a copy of a known virus but a substantially different sequence that still worked.
Engadget covered the same development under a blunter framing: "AI is now making new viruses," with the subheading "What could possibly go wrong?" That skepticism points at the obvious question. The same capability that lets a model design a virus to kill harmful bacteria — a long-standing goal in medicine, since bacteria-killing viruses are studied as an alternative to antibiotics — is a capability that does not, on its own, distinguish between helpful and harmful targets.
That is where the policy problem sits. A report circulated as "AI and Biosecurity Risks Outpace Federal Regulatory Response," published by Legis1 and surfaced via Google News, argues exactly what its title says: the pace of AI-driven biological capability is running ahead of the federal government's ability to write and enforce rules for it.
It is worth being precise about what these sources do and do not establish. They describe published research using genome models to generate novel viral sequences, and a policy analysis warning that oversight lags. They do not document a release, an accident, or harm.
Why it matters: the tools to design living things are now improving faster than the rules meant to govern them, and the window to set those rules is closing while the science is still in the lab.