MI5 Director General Sir Ken McCallum has warned that artificial intelligence could significantly alter the threat posed by existing bioweapons and enable the creation of pathogens not previously seen. He made the warning during his annual assessment of threats facing the United Kingdom, describing AI-assisted biological weapons as a risk that has not yet materialized but warrants urgent attention.
McCallum identified two distinct concerns. AI could make dangerous biological knowledge easier to obtain for people without specialist training, while more advanced systems could help capable organizations design novel pathogens. He said terrorist groups would like access to chemical, biological or radiological weapons, but obtaining the materials, facilities and expertise remains difficult.
The warning came as MI5 described a broader shift in the UK security environment. The agency said its work countering state threats has quadrupled over the past decade, while terrorism investigations remain near record levels. McCallum also said hostile actors have used AI for reconnaissance and attacks, including drones, and that the technology is intensifying the contest between cyber attackers and defenders.
The UK has established an AI Threat Assessment Center to examine emerging dangers. McCallum said the possibility of an autonomous system that could outthink humans represents a new kind of security problem, although he did not claim that such a system has already produced a biological weapon.
The biological risk is not limited to inexperienced individuals. AI tools could help well-funded states or organizations with trained scientists, high-containment laboratories and access to equipment identify genetic changes to existing pathogens. A more distant possibility is the design of entirely new pathogens or toxins that resist drugs or lack effective countermeasures.
Tom Inglesby, director of the Johns Hopkins Center for Health Security, has called for independent safety evaluations of advanced AI models before release, restrictions on sensitive biological data and stronger screening of orders for synthetic genetic material. Existing screening is largely voluntary, and newly designed sequences may evade systems that check only known dangerous genetic signatures.
Significant barriers remain. Producing a biological weapon requires more than written instructions: would-be developers still need tacit laboratory expertise, biological materials, specialized equipment, experimental validation, manufacturing capacity and a workable delivery method. The same AI capabilities can also support legitimate research, including the development of medicines, vaccines and antidotes, making malicious use difficult to separate from beneficial work.
Recent incidents have nevertheless sharpened official concern. An exercise involving non-science students found that chatbots could provide information about potential pandemic pathogens and synthetic DNA suppliers. Anthropic has also described cases in which scientists attempted to use its chatbot for work involving viruses or toxins, including efforts to plan mutations in a highly pathogenic bird-flu virus. Those examples demonstrate access to potentially dangerous guidance, but they do not establish that an AI system has independently created or released a human pathogen.