In an exclusive study, leading experts have warned that recent developments in artificial intelligence could make man-made pandemics five times more likely than they were just a year ago, raising serious concerns about our biosecurity.
Introduction: A new threat on the horizon
As the world continues to recover from the COVID-19 pandemic, a new study shared exclusively with TIME magazine by the Forecasting Research Institute reveals a new and alarming threat. According to the study, recent advancements in artificial intelligence could increase the likelihood of man-made pandemics by fivefold compared to last year. This warning echoes concerns raised by leading AI companies such as OpenAI and Anthropic, which have warned that current AI tools are nearing the capability to assist bad actors in creating biological weapons.
How can artificial intelligence help create a biological weapon?
Biologists have long been able to modify viruses using laboratory technology. But what is new is the ability of chatbots—such as ChatGPT or Claude—to provide precise troubleshooting advice to amateur biologists attempting to create a lethal biological weapon in a lab. Seth Donahue, one of the study’s authors, says safety experts previously considered the difficulty of this troubleshooting process to be a major barrier preventing terrorist groups from creating a biological weapon. He adds: “Now, thanks to AI, the expertise needed to intentionally cause a new pandemic could become accessible to many, many more people.”
Alarming study findings
Between December 2024 and February 2025, the Forecasting Research Institute asked 46 biosecurity experts and 22 “superforecasters” (individuals with a proven track record in predicting future events) to estimate the risk of a man-made pandemic. The median prediction among survey participants for the risk of this happening in any given year was 0.3%.
Importantly, researchers asked another question: How much would this risk increase if AI tools could match the performance of a team of experts in a difficult virology troubleshooting test? If AI could do that, the average expert said, the annual risk would jump to 1.5%—a fivefold increase.
What the forecasters didn’t know was that Donahue’s team was already testing AI systems for this very capability. In April, the team revealed the results: today’s leading AI systems can outperform PhD virologists in a difficult troubleshooting test.
Can these risks be mitigated?
Fortunately, the study also identified ways to reduce the biological risks posed by AI. These preventive measures fall into two main categories:
1. Model-level safeguards: Researchers welcomed efforts by companies like OpenAI and Anthropic to prevent their systems from responding to prompts aimed at building a biological weapon. The study also outlines that restricting the spread of “open-weights” models and adding defenses against model jailbreaking would likely reduce the risk.
2. Imposing restrictions on DNA synthesis companies: Currently, it is possible to send a genetic code to one of these companies and receive biological materials matching that code. These companies are not legally required to screen the genetic codes they receive. The study’s authors recommend that laboratories screen their genetic sequences for harm and implement “know your customer” procedures.
Conclusion: A new risk requiring a swift response
If all these safeguards are implemented, the risk of an AI-driven pandemic could drop to 0.4%, according to median expert forecasts. Josh Rosenberg, CEO of the Forecasting Research Institute, says: “Overall, this appears to be a new risk area that deserves attention. But there are good policy responses to it.”
The message is clear: while artificial intelligence offers tremendous promise, it also brings unprecedented risks. Navigating this new landscape requires a careful balance between encouraging innovation and imposing robust safeguards to protect humanity’s future from potential biological threats.
