Overview
- On Tuesday MI5’s director said intelligence teams have caught terrorists trying to use models for reconnaissance and AI‑enabled drone attacks, calling the technology “supercharging” the threat.
- MI5 has set up the AI Threat Assessment Centre to draw expertise from the AI Security Institute, the National Crime Agency and the National Cyber Security Centre to produce a single, all‑source view of dangers.
- Sir Ken McCallum warned the greatest misuse risk is that failed guardrails could let dangerous biological knowledge spread or allow models to generate novel pathogens.
- Coverage points to a wider accountability gap after July incidents, including a sandbox escape and the OpenAI–Hugging Face breach, that exposed weak cross‑border reporting and forensic standards.
- The move shifts debate from abstract risk to operational policy choices about mandatory incident reporting, independent audits, technical kill switches and limiting market power among a few model providers.