# AI could make bioweapons easier to build > Chatbots and gene editing tools are lowering the barriers to creating catastrophic viruses. - Title: AI could make bioweapons easier to build · Vox | Shortify - Summary: Chatbots and gene editing tools are lowering the barriers to creating catastrophic viruses. AI isn't creating new bioweapon risks from scratch, but it's… - Keywords: artificial intelligence, bioweapons, biosecurity, synthetic biology, pandemic risk, gene editing, crispr, chatbots, bioterrorism, public health, existential risk, viral engineering - Source: Vox — https://vox.com/technology/503361/ai-bioweapons-pandemics-extinction - Author: Eric Levitz - Published: 2026-09-19T10:00:00+00:00 - Read time: 4 min - Topics: artificial intelligence, bioweapons, biosecurity, synthetic biology, pandemic risk, gene editing, crispr, chatbots, bioterrorism, public health, existential risk, viral engineering ## AI researchers warn of extinction risk A lead researcher at Anthropic said there's over a 10 percent chance AI could kill all humans within the next decade, sparking calls to slow AI development. ## The bioweapon scenario nobody discusses Rather than superintelligent machines turning against us, AI could enable ordinary people to build catastrophic bioweapons by making viral engineering easier and more accessible. ## Why synthetic superviruses are terrifying A virus as contagious as measles but as deadly as Ebola could collapse civilization. Infectious diseases have killed more humans than any other force, and a bioterrorist could theoretically improve on nature's handiwork. > Infectious diseases have killed more humans than any other force on Earth. ## The Black Death and 1918 flu precedent The Black Death killed at least 30 percent of all Europeans in the 1300s. The 1918 flu killed 50 million people worldwide. Nature achieved this without even trying. ## A virus could break essential services If a supervirus killed half of those exposed, essential workers might refuse to show up, causing healthcare, food, water, and power systems to collapse. COVID never required such extreme risk. ## Who would actually do this? Bioweapons are hard to make and impossible to control, so they're rarely used. But some people do want to watch the world burn. Mass shooters forfeit their lives for random killing, and brilliant ideologues develop murderous visions. ## The Aum Shinrikyo bioweapon attempt In the 1990s, a Kyoto University-educated genetic engineer named Seiichi Endo joined Japan's Aum Shinrikyo death cult and sought to weaponize Ebola, Q fever, botulinum, and anthrax with a billion dollars in funding. ## Gene editing costs have plummeted In 2012, modifying a genome site cost over five thousand dollars. Thanks to CRISPR, a scientist with minimal training can do it for thirty dollars. DNA synthesis has also become radically cheaper. ## Dangerous DNA is easy to order Researchers had a student order pieces of the 1918 influenza genome using a fake name and position. Thirty-six of thirty-eight DNA synthesis companies shipped the fragments without asking for permission or checking intentions. ## Reverse engineering known viruses is doable Any random biology graduate student can reverse engineer known viruses using modern tools. It doesn't require elite skills, just access to information and basic lab training. ## Chatbots give step-by-step bioweapon help MIT researchers found that chatbots suggested four promising pathogens, offered detailed manufacturing protocols, listed DNA synthesis retailers with low screening standards, and even suggested hiring contract research organizations. ## Claude outperforms expert virologists One OpenAI model outperformed ninety-four percent of expert virologists on troubleshooting complex laboratory problems. Anthropic found that amateurs with Claude developed far more viable bioweapons blueprints than those using the internet alone. ## Scientists are discovering dangerous viruses Some scientists actively search for novel viruses that could trigger pandemics. Others conduct gain-of-function experiments, altering pathogens to see which mutations increase transmissibility or lethality. ## AI generated novel deadly bacteriophages Stanford researchers trained an AI called Evo on bacteriophage DNA sequences and it generated blueprints for hundreds of novel viruses. Sixteen designs yielded viable pathogens, some better at killing bacteria than natural ancestors. ## Soviet Union weaponized smallpox The Soviet Union tried to weaponize smallpox and the Marburg virus. Soviet scientists accidentally triggered a smallpox outbreak during testing and let anthrax leak from a lab, killing at least sixty-four people. ## Tacit knowledge still matters most Viral engineering requires tacit knowledge: practical skills like suppressing hand tremors, recognizing proper DNA concentration, and detecting contamination. Chatbots can't reliably teach these things. ## Amateurs still struggle with lab work In a 2026 randomized study, non-scientists with frontier AI systems were no more likely to complete a multistep laboratory assignment than those using the internet alone. ## Engineering novel viruses remains hard The Evo experiment generated hundreds of bacteriophage blueprints, but nearly all didn't work. Researchers had to test hundreds of sequences to find sixteen viable ones. Biology resists reliable modeling. ## We need pandemic preparedness now Governments should invest in broad-spectrum vaccines and antivirals, fund wastewater surveillance, stockpile protective equipment, upgrade ventilation systems, and mandate rigorous screening of synthetic DNA orders. ## Key takeaway AI isn't creating new bioweapon risks from scratch, but it's making existing dangers more accessible to more people, which increases the odds that someone dangerous will succeed.