Can AI create a dangerous virus? New safety report offers a reality check

New Delhi: Artificial intelligence is becoming surprisingly good at helping scientists with their difficult tasks, including work related to virology. As it has become more advanced, experts are concerned that it could be misused in biological research. The 2026 International ‘AI Safety’ Report has taken this issue seriously, with more than 100 independent experts across over 30 countries, and its overall message is cautious rather than alarmist.

The report says that one AI model performed better than 94% of domain experts on tasks involving troubleshooting virology laboratory protocols, which shows how advanced AI has become. However, this does not mean the technology can completely shift to AI on its own, and it also has the potential to create a pandemic-causing virus. Developing a biological design into a real-world threat remains a difficult process. It involves scientific knowledge, laboratory access, specialised materials and other practical challenges. Whether an AI-assisted design could become dangerous is a separate question from how a model can answer technical questions about virology.

What can AI do in biological research?

AI tools can explain difficult scientific concepts, help researchers work through laboratory problems and provide them with information about biological processes. The report also stated that some systems can also produce technical guidance that might be useful to someone trying to develop biological or chemical weapons.

Researchers have explored how AI can help design biological systems. Some of this work has involved bacteriophages – viruses that infect bacteria, not people. Creating a design on a computer, however, does not automatically produce a viable virus, let alone one capable of spreading between people.

Professor Wendy Barclay, Regius Professor of Infectious Disease at Imperial College London, has described AI as a double-edged sword. The technology could help scientists develop vaccines and respond to outbreaks, but the same advances raise questions about how biological information might be misused.

The risk also depends on how it is being studied and how much scientists know about it. Scientists still do not fully know how AI-generated biological designs would behave in real-world systems.

Some AI developers have strengthened their safeguards after safety assessments raised concerns that newer models might help inexperienced users with certain aspects of biological weapons development. The concern is not proof that these systems can create a pandemic, but that their capabilities may be advancing faster than the methods used to assess them.

Experts say safety testing needs to look beyond how a model responds to individual questions. It should also examine whether an AI system can connect information, solve problems across multiple stages of a research process and use tools to complete more complex tasks.

At the same time, naturally occurring pathogens remain a major concern for public health experts. The COVID-19 pandemic shows how fast a new infectious disease can put pressure on the healthcare system and an ordinary person’s life.

A complete ban on virus research is not considered the only answer. Stronger screening of orders for synthetic biological materials, better laboratory oversight and more thorough AI safety evaluations could help reduce the risks without stopping legitimate scientific research.

For now, there is no reliable estimate of the likelihood that AI will trigger the next pandemic. The challenge is to understand what these systems can realistically do, identify gaps in existing safeguards and strengthen oversight as their capabilities develop.

Written By Uditi Sharma!