In the mid-1990s boy scout David Hahn used household objects and his scientific knowledge to start building a nuclear reactor in his backyard. Police and the Environmental Protection Agency stopped him before he could finish. Twenty years later, revolutions in manufacturing and computing have made projects such as Hahn’s a lot more feasible; if he had access to a 3-D printer, for example, he might have finished his reactor before authorities intervened. Modern technologies also mean one does not need to be as smart as Hahn to create at least some kinds of DIY weapons. With the right machine and blueprints anyone can build a handgun in their living room—and firearms are just the beginning. Researchers fear that artificial intelligence and 3-D printing might one day create, on demand, weapons of mass destruction.

A report published Tuesday from a multi-institutional research group led by the Middlebury Institute of International Studies at Monterey sounds an alarm about this possibility. “This is the proverbial wicked problem,” says paper co-author Robert Shaw, Director of the institute’s Export Control and Nonproliferation Program. Shaw says the proliferation of 3-D printers, combined with advances in artificial intelligence, could make it much easier for nations or individuals to covertly build nuclear, chemical and biological weapons.

When most of us think of 3-D printing, also known as additive manufacturing, we picture jets depositing layers of plastic to create models for hobbyists. But the potential for the technology goes much further—some researchers have claimed that the ability to print almost anything imaginable will usher in a new industrial revolution.

Industrial-scale 3-D printers are already advancing technology related to extremely dangerous weapons. The U.S. National Nuclear Security Administration (NNSA), for instance, is using these machines to manufacture models of nuclear weapons for testing. “While using 3-D printing to maintain the nation’s nuclear weapons stockpile, NNSA labs are advancing the broader science of the field,” the administration said in a 2016 blog post. Meanwhile defense contractor Raytheon has a 3-D printer that can manufacture 80 percent of a missile, and Los Alamos National Laboratory is using these machines to produce high explosives.

Not even an advanced machine can print weapon components without expert designs—but there are concerns that artificial intelligence could allow unskilled humans to come up with the necessary blueprints, thanks to a technique called generative design. With this process a user can give a computer a design problem and set requirements for the final result. The AI suggests many possible solutions, and humans pare down the results. For example, when General Motors wanted to replace its heavy eight-component seat belt bracket with a lighter version that used fewer pieces, engineers plugged these parameters into a generative design algorithm and 3-D printed the result as a solid piece of material. NASA recently used the technique to create a prototype lander for space missions.

“The skill that used to be a barrier for developing enrichment equipment for nuclear weapons might be taken out of the hands of a human,” says Ferenc Dalnoki-Veress, a Middlebury physicist who co-authored the report. “This is potentially dangerous.”

The combination of additive manufacturing and AI creates three major risk areas, according to the report’s authors. The first would involve a nation with an existing missile program, such as North Korea, improving its output by printing rocket parts or propulsion-system components. The use of 3-D printers could “advance its capabilities, accelerate its capabilities, or augment its production capabilities in a way where they can produce more missiles more quickly, or at least prototype missiles more quickly,” Shaw says.

The second potential danger is that 3-D printers could help establish a weapons program by producing the required infrastructure—without alerting international watchdogs. Observers can currently monitor the global supply chain for signs someone is building a factory meant to produce weapons of mass destruction; this is one reason the import and export of certain substances, such as ammonia-based fertilizer (a key component in homemade bombs), are tightly regulated and scrutinized. But industrial 3-D printing could potentially bypass some of the world’s existing arms control frameworks. Today if a country wants to manufacture a dangerous chemical such as sarin, it needs to go through public channels to purchase a particular kind of noncorrosive metal piping. Soon it may have the ability to print these supplies instead.

The third risk is a “black swan” event, a threat no one sees coming. “We see the possibility we could have something completely new, that no one here is really thinking of, that could have weapons of mass destruction capabilities,” Shaw says.

Not everyone is convinced 3-D printing and generative design will ramp up the possibility of apocalypse. “For state actors, 3-D printing can be useful … but I don’t see that additive manufacturing magically solves a lot of problems,” says Martin Pfeiffer, a doctoral candidate at the University of New Mexico and an expert on the anthropology of nuclear war. For nonstate actors, he adds, “additive manufacturing could let you do some things with a smaller visibility footprint, but you can’t 3-D print a plutonium or [highly enriched uranium] core.”

Giacomo Persi Paoli, a researcher at nonprofit research institute RAND Europe who investigated 3-D-printed small arms for the United Nations, sees parallels between his work and that of the Middlebury researchers. “What they’re saying is plausible,” he says, but he notes that successfully 3-D printing a weapon is harder than it sounds. “It’s a combination of four things,” he explains. Making a weapon requires a digital blueprint, the 3-D printer itself, the material that will be shaped by the printer, and human labor to finish the weapon. “It’s very unlikely that whatever comes out of the printer will be plug-and-play, ready to go,” he says.

Take the Liberator, one of the earliest functional 3-D printed handguns. It was the result of hundreds of iterations and a lot of human effort. The completed design did not come out of the printer ready to shoot: it required a careful assembly, and even then the gun did not work all the time. The same would likely be true of more complex weapons. Generative design promises to make parts of the process easier, but consider the instances in which generative design produced good results: GM’s seat belt bracket and NASA’s lander. Both projects had engineers shaping the design, not amateurs.

That does not mean these worries can be dismissed altogether. “The genie’s out of the bottle—it keeps me up at night,” says Gretchen Hund, former director of the Pacific Northwest National Laboratory’s Center for Global Security. Hund, who built a career on identifying emerging issues and technologies that might have national security implications, says the additive manufacturing market has exploded in the past five years—and remains largely unregulated.

“Given the speed at which technology is advancing ... this is something policymakers should be aware of,” Paoli says. “This is a plausible threat. It’s not immediate, but it shouldn’t be discounted.”

The new report suggests that 3-D printing could be part of the WMD supply chain within the next 10 years. “Right now people aren’t paying enough attention to it,” says co-author Miles Pomper, a senior fellow of the James Martin Center for Nonproliferation Studies. “And this is trying to sound that alarm to get them to focus.”