The head of the United Kingdom’s domestic intelligence service, MI5, has highlighted the potential for artificial intelligence (AI) to significantly impact the development of biological weapons, warning that AI could “change the stakes” in this domain. This concern was outlined during Sir Ken McCallum’s annual national security briefing, underscoring evolving threats posed by emerging technologies.
AI’s role in bioweapons development varies depending on the actors involved. For less skilled individuals or terrorist groups operating outside of formal laboratories, AI chatbots could assist by aggregating relevant information, guiding experiment planning, and troubleshooting technical issues, primarily focusing on weaponizing existing pathogens or toxins. On the other hand, state-backed organizations and well-resourced groups with access to high-containment labs and trained personnel stand to benefit more immediately. AI tools can enhance their ability to genetically modify pathogens, potentially increasing their lethality or resistance to treatments. Experts note that AI amplifies existing expertise rather than replacing it.
There is also a longer-term concern that highly sophisticated groups may eventually use AI to engineer entirely new pathogens designed to resist current drugs or evade all known antidotes.
Recent incidents have brought these issues into sharper focus. In 2025, Indian authorities uncovered a suspected bioterrorism plot in which a doctor and associates reportedly sought AI-generated guidance on producing ricin, a lethal toxin, using improvised kitchen equipment. While the effectiveness of the AI responses is uncertain, the incident illustrates how such tools could facilitate dangerous activities. Similarly, the AI firm Anthropic reported multiple instances earlier this year where scientists from undisclosed countries attempted to use its Claude chatbot to advance work related to highly pathogenic viruses and toxins. In one highlighted case, the chatbot was used to plan experiments introducing mutations into a bird flu virus to enhance its transmissibility in mammals. Although these studies carry dual-use implications, similar research is conducted legitimately to anticipate and prevent pandemics.
Further research has demonstrated how AI can provide potentially hazardous information. An exercise conducted at a U.S. university showed that even individuals without scientific backgrounds could coax chatbots into detailing how to synthesize pandemic-causing agents and identify DNA synthesis companies with lax screening protocols. Other examples include AI outlining methods for pathogen modification to evade treatments and proposing airborne release mechanisms using weather balloons.
The dual-use nature of biological science complicates efforts to balance security concerns with beneficial research. Work conducted at Los Alamos National Laboratory on neurotoxins from marine snails, aimed at developing antidotes, reportedly inspired parallel efforts by Chinese researchers to create AI tools for toxin design, which the Chinese team described as purely pharmaceutical. Additionally, scientists have recently used AI to design novel viruses targeting bacteria, a development with promising medical applications but prompting calls for enhanced biosafety and biosecurity measures.
Despite these risks, significant barriers remain to successful bioweapons development, particularly for unskilled actors. Beyond access to information, producing biological weapons requires tacit expertise, specialized materials, validated laboratory experiments, and methods for scaling and delivery. Experts emphasize that AI cannot substitute the unwritten knowledge and resources historically critical to state-run bioweapons programs. Physical access to biological materials and containment facilities remains a major limiting factor.
At present, few AI companies publicly disclose comprehensive strategies to prevent misuse of their tools for biological weapons development. Security experts advocate for formal government policies, mandatory independent risk assessments of AI models before release, and tighter regulation of high-risk biological data. They also stress the importance of rigorous screening of genetic material orders by synthesis companies to prevent the manufacture of dangerous pathogens. However, such screening is largely voluntary and can be circumvented when novel genetic sequences, potentially designed with AI assistance, fall outside existing monitoring frameworks.
As AI technologies continue to advance, officials and researchers emphasize the need for proactive measures to mitigate bioweapons risks while preserving the benefits of biological research and innovation.