When people talk about the dangers of artificial intelligence, the conversation usually revolves around replacing jobs, the possibility of AI taking over the world or the dangers of data leaks and deepfakes. But a new threat that has begun to raise concerns is the use of AI to assist in developing technologies that pose a much greater threat to humanity: bioweapons.  

The United States is no stranger to bioweapons. During World War II, testing began on several dangerous bioengineered microorganisms. The original intent was to develop defenses against potential Axis bioweapons, but research soon took a turn for the offensive to gain a tactical advantage. This research continued into the Cold War with the development and testing of the bioterrorist agent Bacillus anthracis, or anthrax. While the development and stockpiling of these weapons have long since been outlawed, concerns surrounding bioweapons have resurfaced in relation to recent AI-centered research. 

Anthropic, the company behind the AI giant Claude, recently stated that they have had to intervene and disrupt several attempts to use artificial intelligence models to develop research that could aid in the development of biological weapons. An Anthropic report about the misuses of its Claude AI model disclosed several attempts to use this AI to develop software for firearms, missiles, armed drones and bombs in China, Russia and Yemen, as well as several biological research cases involving undisclosed institutions possibly being used for the development of biological weapons.  

While the intent of these research institutions is unclear, it is known that scientists circumvented controls and safeguards to conduct their research. In a case from May, a scientist was found to be studying a harmful mosquito-borne virus called the chikungunya virus. Their research focused on how the virus could be engineered and mutated to become more harmful through repeated infections of living organisms.  

This case with Anthropic shows that artificial intelligence does not need to be a weapon itself to pose a serious threat. If it can conduct potentially dangerous research more quickly and make that research accessible, then the risk comes from what people can and will do with it. As AI continues to advance, governments and companies must develop stronger safeguards and regulations as quickly as these models develop. The real threat is no longer only what artificial intelligence can do, but also what it can teach humans to do.