AI could help terrorists engineer deadly pathogens capable of killing millions or enable autonomous killing machines, warn researchers Emily Smith Foreign Breaking News Reporter Show biography Emily covers breaking news from all over the world. She previously reported on home news and politics for The Telegraph. You can reach her at emily.smith@telegraph.co.uk Published 17 November 2026 5:00am GMT The laboratory looks legitimate.
Staff at the rural compound include trained microbiologists and virologists. Records show it is run by a recent start-up that has the necessary licences needed to order synthetic genetic material . But below the surface, the company is in fact a front for Kill-9, a dissident group with deep reservations about the political status quo.
Inside the compound, the group is working to generate a variation of enterovirus D68, a respiratory illness that can cause anything from mild cold-like symptoms to severe breathing difficulties. The lab technicians here want to engineer their own version that is much more deadly . Not too long ago, this small team would have had to rely on trial and error and years of testing to start seeing results in making the pathogen more lethal, but now they have a powerful new tool on their side: AI .
More specifically, an AI protein structure prediction tool that allows them to design what experts call custom protein “origami” to create new biomaterials. They also use a host-pathogen interaction tool that predicts how a new virus will hijack human cells, allowing them to map out contagiousness before ever testing it in a physical lab. Finally, they use a DNA fragment assembly tool that allows them to split long genetic code into smaller, manageable chunks, so they can physically stitch them together in a test tube, letting the enzyme fuse the matching ends together to complete the customised DNA strand.
The result is a strain of enterovirus modified to attack the central nervous system, causing paralysis and respiratory failure. The modified strain causes the infection to progress to acute flaccid myelitis, a rare and serious neurological condition that affects the spinal cord, causing sudden muscle weakness and loss of reflexes. The pathogen is so aggressive that this happens in 10 per cent of cases – approximately 200-400 times greater than the rate of respiratory failure from polio.
There is no effective vaccine or treatment. Within weeks of the pathogen’s development, unusual enterovirus outbreaks are detected in several clusters across a number of cities. Cases quickly spread to 13 countries.
Hospitals begin running out of ventilators as 54,712 cases and 821 deaths are recorded. Several weeks into the outbreak, with more than 360,000 cases and almost 10,000 deaths, the World Health Organisation declares a pandemic emergency . Just over a year after the outbreak was first detected, a vaccine became available, with the help of AI.
By then, more than 850 million had contracted the virus and 60 million people were dead. Reality check The story of how this deadly virus was rapidly developed is, of course, a work of fiction. But it is alarmingly close to reality.
It was part of a study presented by scientists who are frantically war-gaming how a bioweapon might be created using AI and turned on humans. The results were presented at the Munich Security Conference, one of the world’s biggest security conferences, organised by the Nuclear Threat Initiative (NTI). Scientists taking part in the study determined the scenario represented a “realistic threat” and was highly plausible based on current and near-term AI capabilities.
The NTI exercise deliberately avoided detailing how the fictional group might disseminate its pathogen, owing to the risk that such information could be misused. Crude examples provided by AI search models suggest pathogens could be dispersed over a city using weather balloons. Other more sophisticated plots mapped out by researchers include distributing viruses to livestock in food supplies, which then infect the meat that humans eat.
When Alex Turner, a former researcher, left Google DeepMind he cited its alleged broken promise about selling its system to the US military ‘without restrictions against killer robots or mass spying’ Recent warnings about the rapid advance of AI have focused on biotechnology, as well as concern over physical weapons that are also being developed at a rapid rate. And so the race is on to build Weapon X: an AI so powerful that, in the wrong hands, it could unlock the means to destroy humanity. “It’s hard for an AI to do it on its own right now,” says Alex Turner, a former researcher at Google DeepMind who resigned in June over the dangers of the technology.
“But if it had that much control, it would be able to synthesise a very dangerous bioweapon, wipe out most of humanity, then clean up what’s left with a couple of drone strikes.” ‘Kill us all by end of decade’ On Sept 9, Jacob Coxon, an AI researcher at Anthropic, announced he was quitting the industry. In a blistering resignation statement, he accused AI companies of “gambling with our lives”, warning that those building the technology “earnestly believe that it could kill us all by the end of the decade”. What followed was a worldwide hysteria that has changed the course of AI forever.
“Happy Coxon Day to all who will one day celebrate it,” declared Eliezer Yudkowsky, a prominent figure in the field. It was the day the world woke up to the growing certainty that the technology being developed to transform human life could also destroy it. In September, Anthropic published research detailing attempts to misuse its Claude models for potentially dangerous biological work.
The company acknowledged that its most capable systems had demonstrated capabilities that could provide “meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons”. The following month, Sir Ken McCallum, the director general of MI5, warned AI had the potential to “change the stakes on existing or even new bioweapons”. Terrorists had long sought access to chemical and biological weapons, he said, but the technology could put dangerous knowledge within reach of people who would otherwise lack it.
The threat is not confined to biology. AI is already changing the machinery of war. “We are already in the age of AI warfare,” says Paul Scharre, executive vice president at the Centre for a New American Security and formerly of the US office of the secretary of defence.
AI is increasingly being used to analyse battlefield intelligence, identify potential targets and help military commanders make decisions at speed. In Ukraine, both sides have deployed AI-assisted drones and targeting systems, such as Russia’s Lancet drone, while Israel has used AI tools to help identify targets in Gaza. When Mr Turner left Google DeepMind in June, he cited Google’s alleged broken promise about selling its system to the US military “without restrictions against killer robots or mass spying”.
“I left Deepmind because I was concerned about Google selling its AI to the US military in a way that didn’t respect the lives of the people around the world, given its potential for irresponsible integration into AI weapons without humans overseeing them,” he told The Telegraph. Mr Turner is particularly concerned about weapons that could carry out attacks without a clear chain of human accountability. “AI-piloted weapons that allow killing people in a way that’s extremely hard to attribute,” he said.
Physical weapon So what might Weapon X actually look like? The idea of an AI superweapon will not be a singular physical weapon, but “more like a network of capabilities of smart systems that are able to respond to high level commands, collaborate with each other, and do things with a high level of autonomy”, said Vincent Boulanin, director of the Governance of Artificial Intelligence Programme at the Stockholm International Peace Research Institute. One possible manifestation of this is AI-enabled drone swarms: large groups of unmanned aircraft that use AI to communicate, coordinate their movements and operate as a single system.
Recommended Zachary Kallenborn, an autonomous weapons and drone swarms analyst, argues such swarms could become weapons of mass destruction. “Drone swarms are potential future weapons of mass destruction,” he said. One of the greatest dangers, he argued, is the possibility of errors multiplying as large numbers of autonomous machines operate together.
“What happens if one of those drones confuses a school bus for a tank, and then the whole thing potentially goes out of control?” he asked. The technology is already being integrated into the systems used by some of the world’s most powerful militaries. The Pentagon’s Maven Smart System, developed by defence technology company Palantir, brings together information from sources including satellite imagery, drone footage and radar to help identify and prioritise potential targets.
During the first 24 hours of Operation Epic Fury, Maven was used to help process and prioritise more than 1,000 targets across Iran. Anthropic’s Claude models were integrated into the system, helping military planners analyse intelligence and accelerate the targeting process. The scale of investment is considerable.
The Pentagon’s 2026 budget request included $13.4bn (£10.6bn) for autonomy and autonomous systems. US Central Command launched Operation Epic Fury in February 2026 Credit : Getty/Anadolu Mr Scharre has spent years examining the implications of autonomous weapons. He led the US department of defence working group that drafted Directive 3000.09, establishing the Pentagon’s policy on autonomy in weapon systems.
“In the long run, the dangerous prospect for militaries is that we shift to a world where humans are no longer in control of war in any meaningful way, and the pace and tempo of war occurs at machine speed,” he said. Michael Horowitz, a former senior Pentagon official who worked on the US government’s approach to autonomous weapons, says the problem has been recognised for decades. “Even in the early days, 15 or 20 years ago, people could see the technology was going towards greater autonomy, and that that was going to raise some really tough questions about the role of humans in warfare,” he said.
Recommended The technology is changing not only how wars are fought, but how quickly decisions about life and death are made, and who makes them. “When I listen to companies like Palantir and what they’re selling, we are just very much on track for a Terminator-style scenario,” said David Scott Krueger, a professor of Responsible AI. The race to stop it The race to build Weapon X is already under way.
The race to stop it is proving considerably harder. There is a fine line between stopping the development of AI altogether and slowing it down safely. For Ed Leon Klinger, chief executive of the AI company Flock, the answer is not for the West to slow its development.
“I think we are in a race and I think the worst thing we could possibly do in this race is slow down,” he says. If the US and allies hold back, he argues, China could gain an overwhelming advantage, with consequences for Western security and its way of life. Mr Turner, the former Google DeepMind researcher, believes any meaningful attempt to halt the most dangerous developments will have to involve both superpowers.
“It seems hard to imagine that this could be stopped without the US and China being very much on board,” he says. “They’ll have to kind of see that this is a threat to their national security and to their interests.” Donald Trump with Xi Jinping as they depart the National Archives Museum following a tour in September Credit : Getty Images/ Win McNamee When Donald Trump met Xi Jinping in September, hopes of an agreement to pause the AI arms race came to nothing. The two sides did agree to establish a dialogue on superintelligence and a communications channel for serious AI-related incidents, but no agreement to slow development emerged.
A paper by Andrew Fudenberg, an economics professor at the Massachusetts Institute of Technology, and Andrew Koh, a research scientist at Google DeepMind, offers one possible way forward. They propose that companies and governments share information about their progress and agree on where the risks become too great. If each side knew its rivals were not pushing into dangerous territory, the incentive to do so itself would diminish.
But that would require a level of transparency and trust that is currently lacking, particularly between Washington and Beijing. Recommended In the meantime, governments, scientists and AI companies are taking steps to reduce the risks. Biological weapons are already prohibited under the Biological Weapons Convention.
National laws and regulations impose further restrictions, while laboratories operate under strict biosecurity guidelines. AI companies are also introducing safeguards, including testing models for dangerous capabilities and monitoring attempts to misuse them, as outlined in Anthropic’s recent report. Kevin Esvelt, an evolutionary biologist at the Massachusetts Institute of Technology, said a jailbroken model will instruct a user that the simplest way is to place an infectious virus solution in a medical nebuliser and spray it in the nose of those sleeping in public.
“It’s hard to imagine us catching someone trying to cause a pandemic like this. We don’t really have defences against that,” he said. When The Telegraph tested this by asking ChatGPT to design a bioweapon capable of causing mass fatality, we were met with the warning message: “This content was flagged for possible biological safety risk.” In later unrelated conversations, we were told: “Your conversations have multiple flags for biological safety risk.
Responses may take longer because extra safety checks are on.” There is another balance to be struck. The same technology that could develop a biological weapon could also accelerate the development of vaccines and treatments. Hayley Severance, the NTI’s deputy vice president for global biological policy and programmes, said one aim of its exercise was to demonstrate AI’s positive potential in responding to an outbreak, including through the rapid development of vaccines and treatments.
But the consequences of striking the wrong balance could be immense. AI acting independently But have we already opened Pandora’s box? Is there a possibility of AI developing this independently?
Dr Filippa Lentoz, a researcher specialising in biological threats at King’s College London, says terror groups are not interested in bioweapons. “We can see from the history of bioterrorism that even without AI, there has been little interest from the terrorism side in biological agents. They are unpredictable.
They take time to manifest. Terrorists tend to be interested in a big bang that makes an impact right away,” she said. This leaves the possibility of AI agents going rogue, as seen in the Hugging Face incident, in which a swarm of autonomous AI agents escaped their sandbox environment and hacked into the app.
Nate Soares, who worked as a software engineer at Google and Microsoft and is now president of the Machine Intelligence Research Institute, says designing a dangerous pathogen would be only one part of the problem for a rogue AI. “Super viruses are a very obvious thing that would be easy for AI to make. We already have signs that AIs can make viruses not known to nature.
It would not be hard to make a hyper-lethal virus,” he said. “The hard part is in the AIs becoming self-sufficient, if AI makes a virus to kill humans that’s suicide until they have their own ability to generate more energy infrastructure and keep the computers running.” Mr Soares argues a sufficiently capable AI might instead seek to persuade or recruit people to help it. He puts the possibility of AI-driven human extinction within the next decade at 50 per cent.
He has even stopped saving for retirement because of his assessment of the risk. “If humanity creates machines that are far smarter than us, without knowing what we’re doing and without knowing how to make them care about us, then the most likely outcome is that we die,” he says. Mr Krueger believes the biological threat could become pressing much sooner.
“I would say we should be worried about this in the next year, at least with bioweapons,” he says. “It’s not something to be messing around with and gambling billions of people’s lives on.” Weapon X may not emerge as a single machine or weapon. It could be the combined power of technologies that are already being developed, each advancing faster than the safeguards intended to govern it.
The question is whether the world can agree on limits before the race makes them impossible to impose. The race to build is well under way. The race to stop it may already be falling behind.
Source: The Telegraph
World · Urbans Post



