AI Can Now Create Viruses From Scratch. What Happens When Someone Uses It for Harm?

3 min readHere’s what you’ll learn when you read this story:Earlier this month, the AI company Anthropic detailed a handful of scenarios where potentially bad actors were using its LLM Claude to possibly research and build bioweapons. While experts debate the seriousness of the research involved, the threat assessment reiterated the growing need for regulation around artificial intelligence and synthetic biology. While AI holds immense promise for combating superbugs or curing intractable diseases and cancers, its ability to create novel viruses also makes it an unmistakable bioweapons threat.Any tool is just that—a tool. Yet time and time again throughout human history, our species’ intention ultimately defines the technology, whether a Paleolithic knife or advanced nuclear power, as “good” or “bad.” The rise of artificial intelligence is no different, but the stakes are remarkably higher compared to most tools—and nowhere can this be seen more clearly than in the world of biology.Last month, a team of researchers from Stanford University and the Arc Institute, a nonprofit that works at the intersection of biomedicine and AI, revealed the promising power of Evo 2, a DNA foundation model capable of creating viruses from scratch. To crib a phrase from poet Robert Frost, it’s clear how “two roads diverged in a wood” following this discovery. Novel viruses mean new ways to tackle anti-microbial resistant (AMR) superbugs, which are directly responsible for more than a million deaths every year. The other road? Well, a not-so-short trip down memory lane can conjure a picture of what that might look like. After all, what if instead of developing bacteriophages capable of K.O.’ing E. coli, a bad actor decides a new kind of a bioweapon might serve their devious designs?This growing concern about a new era of AI bioweapons reached a crescendo earlier this month when Anthropic, the AI company behind the popular LLM Claude, released a lengthy threat assessment report detailing a handful of scenarios where users made queries related to potentially dangerous biological research.“These examples are illustrative of the kinds of tasks to which our models are put, and the often-difficult judgements we have to make when assessing whether or not a given biological use is dangerous,” the post reads. “When we detected and investigated these cases, we banned the users’ accounts and incorporated our investigative findings into our frontier model safeguards, enforcement, and threat intelligence processes to better prevent, detect, and disrupt these activities in the future.”Not long after this assessment’s publication, Anthropic CEO Dario Amodei called for a slowdown of AI development, citing “misuse of AI for…bioterrorism” as one of its core reasons (as well as AI’s growing propensity to escape its technological confines). This comes only months after OpenAI and Anthropic signed a letter urging tighter regulations for companies that sell made-to-order DNA strands. Slowly, a deadly equation is taking shape: A tool capable of constructing deadly toxins + methods to produce said toxins = a very dangerous situation.And this threat isn’t just hypothetical. A 2025 study urged the AI model Moremi Bio Agent (without its safety guardrails) to develop novel toxins that closely matched “toxins such as ricin, diphtheria toxin, and disintegrin-based snake venom proteins.” Moremi Bio Agent generated roughly 1020 novel toxic proteins and 5,000 toxic small molecules in total.“This research highlights the urgent need for mitigation and safeguards against the dual-use threat of AI-biodesign tools,” the authors concluded.For now, Anthropic’s bioweapon threat revelation has garnered mixed responses from experts, with some warning that we’re “sleepwalking into a potentially huge disaster,” while others caution that the presented cases represent only basic research, as Science reports. But this summer, more than a few AI close-call scenarios have caught the attention of security experts, AI company C-suites, and even those in the halls of Congress.So far these threats, whether to biosecurity or cybersecurity, have caused little to no damage—but who knows when our luck will finally run out.Darren lives in Portland, has a cat, and writes/edits about sci-fi and how our world works. You can find his previous stuff at Gizmodo and Paste if you look hard enough.