AI could help terrorists build bioweapons: Researchers raise alarm

AI could help terrorists build bioweapons: Researchers raise alarm Artificial intelligence could make it easier for terrorists and inexperienced extremists to obtain the knowledge needed to develop biological weapons, raising fresh concerns among security experts about the rapid growth of powerful AI systems. A recent report has warned that increasingly capable AI models, particularly open-weight systems that can be downloaded and operated privately, could provide dangerous assistance to people who previously lacked the scientific expertise needed for sophisticated biological attacks. Unlike conventional online services, some of these models can be run locally without sending prompts to a company's servers. That could make their use harder for authorities to monitor and could also allow users to bypass safety restrictions built into commercial AI systems. Researchers are particularly concerned about models whose safeguards can be removed or weakened. A recent assessment by Tech Against Terrorism found that almost one-third of responses in its pilot testing provided meaningful assistance that could increase a malicious actor's ability to prepare terrorist activity, despite existing guardrails. The concern comes as AI becomes increasingly capable in biological research. Scientists and policymakers have warned that advanced AI can accelerate tasks ranging from biological research and drug discovery to the design and analysis of biological systems. The same capabilities that could help researchers develop new treatments could potentially be misused for harmful purposes. A RAND report published this month called for a layered approach to the problem, arguing that no single safeguard is likely to prevent AI from being misused to help develop a biological weapon. Its recommendations include restricting access to sensitive information and biological materials, detecting signs of malicious activity and creating stronger deterrents for people attempting to misuse AI.