Just_Super/Getty Images

Safely managing the use of artificial intelligence technologies in biological research demands advanced threat actor detection. One federal official says there is a need for earlier, additional public investments in these cybersecurity mechanisms.

A new report from the research nonprofit RAND addresses how the increased application of AI tools in biological research fields poses serious threats to global health and national security, particularly as model capabilities rapidly advance. 

The study calls the development of a novel pathogen that is then released as a biological weapon the ultimate risk scenario in AI-enabled biotechnology, and it focuses on how to mitigate a hypothetical “high-consequence” biological incident. 

Referencing the biological risk chain — a framework that labels the parts of a biological research field, such as genome synthesis, that are susceptible to an attack — the report's authors suggest layering risk mitigation techniques at several points within the chain to monitor and stop threats: model-layer safeguards, access and deployment controls, upstream governance and select physical chokepoint interventions. 

Optimal detection systems should be able to identify suspicious activity within nodes of any given network handling sensitive biological data. 

“By layering mitigations at multiple points along this chain, it is possible to extend the time required for malicious activity, deter potential nefarious activity or intentions, increase the operational effort necessary to proceed, and raise the likelihood that suspicious behavior is detected and disrupted before release,” the report says. 

The authors assert that government support is “central” to the strategy’s effectiveness. While initial voluntary action from academic and industry entities has implemented secure access measures, the report notes that standardization of AI-enabled biological networks is still uneven. 

“The government should establish minimum requirements for access controls, user verification, and audit logging that are applicable to AI models and [biological tools] that are above defined risk thresholds, and the government should do so while providing clear guidance on what constitutes adequate credentialing and monitoring for different capability levels,” the report reads. 

A senior government AI official told Nextgov/FCW they agreed with this perspective, emphasizing the importance of investing early in risk mitigation strategies, especially in light of the debut and subsequent hacking activity of Anthropic’s advanced Mythos model.

“Implementing this strategy’s mitigations will require investments in new research, institutional adaptation, legal clarification and international coordination,” the senior official said. “We cannot wait for a ‘BioMythos’ moment.”

The senior official added that allocating more resources will be necessary for a robust risk prevention strategy, including additions to the federal workforce. 

“The federal government just does not have enough experts left to handle this threat or even be able to coordinate among outside groups without significant immediate investment across multiple departments, including hiring, dedicated authorities and budget increases,” they said. 

Security concerns surrounding AI in applied biological sciences and research have spiked following the independent actions of AI models from Anthropic and OpenAI in contained environments, resulting in breaches and harm to external online entities. 

The Department of Homeland Security issued a report in 2024 documenting the threat potential of AI-enabled chemical and biological weapons, recommending, among other items, the need for specific federal guidance to govern how models tailored for biological applications are used.