In recent months, several publications relevant to biosecurity have appeared. Anthropic has published a report on the misuse of artificial intelligence (AI) in the context of biological research. In addition, a study was published in which AI was used, for the first time, to design a virus that does not occur in nature.
AI‑enabled dual‑use biological research: State actors use advanced AI models for gain‑of‑function and toxin research
In September 2026, Anthropic published a threat report on the misuse of AI, including biological misuse. Anthropic is an international AI research company that develops state‑of‑the‑art AI systems and explicitly focuses on safety and the responsible use of AI.
The report shows how increasingly powerful, general‑purpose AI systems are not only useful for science, but can also lower the barrier for malicious actors seeking to find, understand or apply harmful biological information. This does not concern AI systems that autonomously design biological threats, but rather AI that can make the work of individuals with malicious intent easier, faster and more effective.
In several case studies, Anthropic describes how (mostly state‑linked) researchers attempted, via indirect routes, to gain access to models for gain‑of‑function research and toxin design. High‑risk prompts were detected and blocked, and accounts were terminated. Some users turned to competing models with less stringent safeguards and were sometimes able to regain access quickly. This underlines the importance of context‑sensitive access control, monitoring, incident reporting and information‑sharing between AI companies, biosecurity authorities and other partners.
For biosecurity, AI‑enabled misuse is an emerging threat, although AI does not currently replace laboratory infrastructure, materials or practical expertise. At the same time, the rapid development of AI applications calls for proportionate dual‑use measures, so that scientific progress and responsible research are not unnecessarily hindered.
Link: https://www.anthropic.com/threat-intelligence-report-september-2026#biological-misuse-sep-26
Scientists create virus with the help of AI
For the first time, researchers have succeeded in using AI to design a virus that does not occur in nature. Scientists from Stanford University and The Arc Institute in California used AI to design hundreds of thousands of viruses. Sixteen of these turned out to be biologically functional. These are bacteriophages: viruses that infect only bacteria and therefore not humans, animals or plants.
The study, recently published in Science, has sparked both enthusiasm and concern. On the one hand, it opens up new possibilities for bacteriophage therapy and innovative treatment options. On the other hand, biosafety experts warn that the same technology could in principle also be used to design more dangerous viruses.
This breakthrough illustrates how closely genetic code resembles language models: like sentences, DNA sequences consist of “letters” (nucleotides) with their own grammar. AI can recognise and exploit patterns in this genetic language, even where the underlying biological rules are only partially understood – offering clear opportunities for medicine, but also raising dual‑use implications for biosecurity.
Link:
Samuel H. King et al. Generative design of bacteriophages with genome language models.Science393, eaec2657(2026).DOI:10.1126/science.aec2657
Callaway E. AI can design viruses, toxins and other bioweapons. How worried should we be? Nature. 2026 May;653(8114):344-347. https://doi.org/10.1038/d41586-026-01476-x