No AI system can give you a credible probability that it will trigger the next pandemic. The 2026 International AI Safety Report, produced with guidance from more than 100 independent experts across over 30 countries, documents real and growing capabilities while maintaining something most viral coverage ignores: honest uncertainty.
Here is the tension worth sitting with. One model evaluated in the report outperformed 94% of domain experts at troubleshooting virology laboratory protocols. Producing a pandemic pathogen, however, still requires physical laboratory access, specialized materials, deep expertise, and successful human transmission.
What AI Can Actually Do in a Lab
The capabilities raising biosecurity concerns are the same ones driving legitimate medical research forward.
Current AI systems can walk users through laboratory procedures and troubleshoot experiments at a level that warrants serious attention. The 2026 report confirms that general-purpose AI can provide biological and chemical weapons-relevant technical guidance, including laboratory instructions. Researchers have also reported genome-scale generative design work involving a modified bacteriophage, though that work involves a virus that infects bacteria, not humans, and a generated design is not a viable, transmissible pathogen.
Professor Wendy Barclay, Regius Professor of Infectious Disease at Imperial College London, describes the technology plainly as a double-edged sword. The same tools supporting vaccine research and outbreak response are the ones raising biosecurity flags, and those two realities cannot be neatly separated.
The potential for harm is going to depend on how comprehensive the knowledge base is about the virus targeted. There’s still a huge amount we don’t know.
Professor Wendy Barclay, Regius Professor of Infectious Disease, Imperial College London
The 2026 report is explicit: substantial physical, technical, and logistical barriers remain between an AI-assisted biological design and an actual pandemic.
Where the Industry Needs to Do Better
AI developers have acknowledged a gap they cannot yet fully quantify, and that admission deserves more scrutiny than it has received.
Multiple leading AI developers added stronger safeguards after their own evaluations found they could not rule out that newer models might meaningfully assist novices with biological-weapons development. Read that carefully: could not rule out. That admission demands ongoing, rigorous public scrutiny. A quiet policy update is not sufficient.
The 2026 report points toward a harder standard that the industry is not yet consistently meeting. Can a system combine information, troubleshoot end-to-end experimental workflows, and operate tools autonomously across a full research pipeline? That is the bar. Developers need to close that gap in their evaluations before a problem surfaces, not after. Barclay offers grounding here: “I think it’s still very unlikely that people are going to be achieving success in AI-designed threats very quickly within a garage lab facility.”
Other infectious-disease specialists, cited by The Atlantic, argue that naturally occurring pathogens remain the more probable source of the next pandemic. Biosecurity attention should not drift too far from conventional preparedness. The COVID-19 pandemic demonstrated, without any algorithm involved, how quickly a naturally emerging virus can overwhelm hospital systems and supply chains.
The path forward is not a blanket ban on virus research, which Barclay notes would damage legitimate science without reliably stopping bad actors. Targeted measures are what the evidence supports: tighter synthesis screening, stronger laboratory oversight, and honest AI evaluations that test real-world end-to-end workflows. AI systems also need to communicate uncertainty accurately, rather than producing confident-sounding answers to questions no one can yet answer. The governance work is genuinely hard, but leaving a documented gap unaddressed is the less defensible choice.
