The UK government has published a revised framework of safety obligations for artificial intelligence developers, introducing binding requirements for transparency and risk assessment across all sectors deploying AI systems. The new rules, announced by the Department for Science, Innovation and Technology, mark a significant shift from the voluntary principles that have guided AI development since 2023.
Under the updated framework, organisations developing or deploying high-capability AI models must conduct mandatory safety testing before release, publish summary reports of risk assessments, and appoint a named senior executive responsible for compliance. The regulations apply to any system trained using computing power above a defined threshold, effectively capturing the most advanced large language models and foundation models currently in development.
“The pace of AI advancement has outstripped the existing regulatory toolkit,” said Dr Sarah Collingwood, director of AI policy at the Alan Turing Institute. “Moving from voluntary to mandatory safeguards brings the UK closer to a regime that can genuinely protect consumers while still encouraging innovation.”
The framework also establishes a new AI Safety Unit within the Department for Science, Innovation and Technology, tasked with monitoring compliance and investigating incidents. Companies found in breach face fines of up to 5 per cent of global turnover or 7.5 million, whichever is higher, with enforcement beginning from April 2027 to allow businesses time to adapt.
Industry reaction has been broadly supportive, though some technology firms have raised concerns about the administrative burden on smaller developers. The TechUK trade association noted that the threshold-based approach would exempt most startups while capturing the largest players, but called for clearer guidance on what constitutes “high-capability” training runs.
The announcement follows similar moves in the European Union, where the AI Act came into force earlier this year. Unlike the EU’s risk-based categorisation system, the UK framework focuses on a single tier of obligations for the most powerful systems, leaving less capable AI to operate under existing sector-specific regulations such as data protection and consumer law.
International observers have described the UK approach as a middle ground between the EU’s comprehensive regulation and the light-touch model favoured by the United States. The government has indicated it will review the framework annually, with the first revision expected in July 2027.
Business groups have welcomed the clarity, with the Confederation of British Industry stating that predictable regulation reduces uncertainty for companies investing in AI research and development. The government has allocated 12 million to help small and medium-sized enterprises understand and comply with the new requirements.