AI safety report finds models can outperform virologists on lab tasks
The 2026 International AI Safety Report shows one model outperformed 94% of domain experts at troubleshooting virology laboratory protocols.
The 2026 International AI Safety Report, produced with guidance from more than 100 independent experts across over 30 countries, documents growing AI capabilities. One model evaluated in the report outperformed 94% of domain experts at troubleshooting virology laboratory protocols.
General-purpose AI can provide technical guidance relevant to biological and chemical weapons, including laboratory instructions. Current AI systems can walk users through laboratory procedures and troubleshoot experiments at a level that warrants serious attention.
Professor Wendy Barclay, Regius Professor of Infectious Disease at Imperial College London, described the technology as a double-edged sword. The same tools supporting vaccine research and outbreak response are the ones raising biosecurity flags.
Producing a pandemic pathogen still requires physical laboratory access, specialized materials, deep expertise, and successful human transmission. Substantial physical, technical, and logistical barriers remain between an AI-assisted biological design and an actual pandemic.
Multiple leading AI developers added stronger safeguards after evaluations found they could not rule out that newer models might meaningfully assist novices with biological-weapons development. The report points toward a harder standard that the industry is not yet consistently meeting.
Barclay said it is still very unlikely that people will achieve success in AI-designed threats quickly within a garage lab facility. Other infectious-disease specialists argue that naturally occurring pathogens remain the more probable source of the next pandemic.
Key facts
- One evaluated AI model outperformed 94% of domain experts at troubleshooting virology laboratory protocols.
- The 2026 International AI Safety Report was produced with guidance from over 100 independent experts across more than 30 countries.
- AI developers added safeguards after they could not rule out that newer models might assist novices with biological-weapons development.
Three perspectives
Neutral
AI capabilities can support both legitimate medical research and the provision of technical guidance relevant to biological weapons.
Positive
Leading AI developers have added stronger safeguards after their own evaluations.
Negative
AI developers have acknowledged a gap in their evaluations that they cannot yet fully quantify.
Your week in news, every Wednesday
The biggest stories of the week for your edition, in your inbox every Wednesday morning, your time. Free, and you can unsubscribe at any time.