UK Labour leaders worry AI power is out of control, Guardian reports

The government is now prioritizing AI security, with MI5 chief Ken McCallum warning that AI is supercharging cybersecurity contests and that extremist groups could use it to unleash bioweapons. A parliamentary evidence session involved executives from Anthropic, OpenAI, DeepMind, and Meta, who were split on whether government testing of new models should be mandatory. While the US administration has reportedly asked companies not to share latest models with the UK’s AI Security Institute, Narayan confirmed cooperation has been retained.

Experts are divided on the primary risks. Geoffrey Irving, former chief scientist at AISI, argues open-source models are already powerful enough to cause serious damage and lack effective guardrails. Conversely, Prof Elena Simperl of King’s College London argues that existential risks are theoretical and that the government should focus on immediate, evidenced threats like hacking and addiction. The UK plans to use its upcoming G20 presidency to push for international regulation, while also investing £400m in AI-focused defence capabilities.