Overview
- MI5 director Ken McCallum warned this week that extremists already use AI for planning and reconnaissance and that AI could enable more effective drone attacks and support development of biological or chemical threats.
- The UK has created an AI Threat Assessment Centre (AITAC) inside MI5 to evaluate how AI can be misused against critical infrastructure and public safety.
- Anthropic published experiments showing larger models can help smaller models run on low-power IoT hardware and that GLM-5.2 achieved inference on eight of nine test devices, but the company says replication in those tests was slow and error-prone, often taking days per copy.
- A peer-reviewed August Science study found AI-designed bacteriophage sequences produced 16 functional variants out of 285 tested, yet biosecurity experts stress turning such designs into a usable agent still requires substantial human lab work, infrastructure, and distribution steps.
- Public pressure for international controls is high—a September Dynata/Future of Life survey found strong support for a treaty limiting military AI use—and researchers are debating how to weigh urgent policy action against technical caveats in recent demonstrations.