On October 13, the UK Parliament’s Business and Trade Committee (covering business, innovation, science and trade) held a hearing on AI safety. Representatives of OpenAI and Anthropic rarely took the step of publicly urging that voluntary industry commitments alone are no longer enough to ensure AI safety and that frontier models should be regulated by binding law.
The key points are as follows:
- Backing a statutory framework: Anthropic’s head for Europe said bluntly that “the rules should be binding,” arguing that voluntary commitments fall far short; an OpenAI vice president also backed a statutory framework and an international convention, with mandatory incident reporting as the first element.
- Pledging an “emergency stop”: The two companies pledged that if MI5, MI6, GCHQ or the AI Security Institute produced evidence that a model was unsafe or being maliciously exploited, they would immediately pull it, pause deployment or pause training.
- A 2:2 split among the four labs: Google DeepMind and Meta declined to endorse new legislation and would only stick with existing standards and regulators, showing the industry is now publicly divided on self-regulation versus legislation.
The UK’s current AI Security Institute testing relies entirely on voluntary company cooperation, and the institute has no power to stop the release of a model deemed unsafe. The hearing could push the UK to become the first country to write independent pre-release testing and incident reporting into law, and to promote globally shared principles through its G20 presidency.
Source: read the original. This text is a summary for reference only.
