On October 13, 2026, representatives from OpenAI and Anthropic appeared before the UK Parliament to advocate for binding regulations on advanced AI systems. The discussion focused on implementing mandatory pre-release safety testing and establishing clear liability frameworks for developers.
The proposed rules would utilize capability-based thresholds, meaning regulatory requirements would scale according to the specific functions of a model. This approach aims to focus obligations on the most advanced systems rather than applying uniform standards to all software.
During the same session, MI5 Director General Sir Ken McCallum warned that AI technologies could be exploited to develop bioweapons. The intelligence agency has established an AI Threat Assessment Centre, signaling that national security concerns may heavily influence future legislative frameworks.
The UK government currently relies on voluntary testing through the AI Security Institute. However, pressure is mounting to fulfill a 2024 Labour Party manifesto pledge to introduce formal legislation, especially following internal safety concerns that led OpenAI to withhold its GPT-6.1 Astra model.
