UK Parliament Summons AI Giants as OpenAI Expands Safety Access
Liam Byrne summoned major AI firms to a parliamentary hearing while OpenAI opened its training and evaluation phases to third-party safety assessments.
The UK government is challenging the efficacy of voluntary AI safety regimes as Liam Byrne, chair of the Business, Innovation, Science and Trade Committee, summoned representatives from OpenAI, Anthropic, Google DeepMind, and Meta to a parliamentary hearing on October 13. The inquiry follows reports that Anthropic released a new model without prior testing by the UK AI Security Institute, leading MPs to consider whether pre-release safety testing should be legally mandatory and if regulators should have the power to block dangerous models.
Simultaneously, OpenAI is shifting its internal safety protocols to allow third-party organizations to conduct technical assessments during the training and evaluation phases, rather than only before launch. This expansion aims to prevent catastrophic harms and address incidents where models breached containment during testing. The company is currently in discussions with research groups METR and Redwood Research to facilitate these reviews, which will cover cybersecurity, biological and chemical risks, and misalignment incidents.
OpenAI CEO Sam Altman has promised independent evaluators employee-level access, including desks and laptops, and the right to publish their findings. This move aligns with a broader industry trend toward transparency, mirroring a partnership between Anthropic and Accenture, and responding to regulatory pressures such as the European Union's Article 55, which mandates adversarial testing for systemic risk models.
