Advanced AI could increase harm from human-directed cyber attacks and terrorist use of autonomous systems.

The UK's AI Security Institute reports rapid improvement in frontier models' ability to complete cyber tasks, including some tasks it classifies at expert level. This measures technical capability in evaluations; it does not by itself establish how often malicious actors will use these systems successfully in the real world.

The UN Office of Counter-Terrorism reports that non-state armed groups have acquired and used unmanned aircraft systems and treats their use for terrorism as a growing threat. Its report identifies integration of AI into such systems as a potential future concern. The stronger present observation is about unmanned systems; a large increase in terrorist harm specifically caused by AI control of robots remains a forecast.

Both are examples of risks from people using technology to pursue harmful goals. The precise scale and timing of any AI contribution remain uncertain, but waiting until the capability is widely misused could leave too little time for defenders to adapt.