Imperial College and Thomson Reuters Launch Five-Year AI Safety Partnership
Imperial College London and Thomson Reuters have launched a formal research partnership focused on foundational challenges in AI safety, reliability, and societal impact. The Thomson Reuters-Imperial Frontier AI Research Lab, announced at Imperial's South Kensington Campus on May 18, 2026, is the first strategic partnership within the university's School of Convergence Science.
The collaboration brings together academic independence with industry expertise at a moment when large-scale AI systems are becoming more powerful and widely deployed across critical sectors. Researchers will tackle three core questions: how to build AI systems that are reliably truthful, how to ensure productivity gains are shared broadly rather than concentrated, and how to maintain meaningful human oversight of increasingly autonomous systems.
Research Focus and Current Work
Professor Alessandra Russo, Director of the Frontier AI Lab, framed the research agenda around interdisciplinary challenges that individual institutions cannot solve alone. The partnership draws on methods from law, computational science, social science, economics, and political science.
The collaboration is already active. Researchers are developing a frontier large language model competitive with leading models, set to launch this summer. The lab is also building a cohort of PhD students who will work in a dedicated facility within Imperial's WestTech London ecosystem, with access to shared computing infrastructure and high-quality real-world data.
Why Models Still Fail at Truthfulness
During a panel discussion at the launch event, researchers identified a fundamental problem: AI models are optimized for likelihood and user approval, not truthfulness. Professor Lucia Specia, Chair in Natural Language Processing at Imperial, explained that models are trained on human preferences to generate outputs people favor, rather than outputs grounded in evidence.
Professor Jonathan Schwarz, Associate Director of the Lab and Head of AI Research at Thomson Reuters, said the solution requires refocusing the training stage. "Models are strongly urged to be helpful and agreeable to humans and to provide a response, even in cases of incomplete or uncertain information," he said. "As a model developer, you have to recognize that sometimes it might be more responsible to not provide a response if the reliability is low."
Legal and Economic Questions
AI's rapid deployment is creating sector-specific challenges. In law, large language models offer efficiency gains and potential access to legal advice for more people, but raise liability questions if the advice is unreliable. Olivia Dhein, Practice GenAI Integration Lead at Baker McKenzie, said the technology creates "an interesting point on access to justice" - expanding opportunity while introducing new risks.
On employment, Kanishka Narayan MP, Minister for AI and Online Safety, said some sectors face significant overlap between AI and human capabilities, while others can be augmented rather than replaced. The UK government is studying AI's effects across sectors to build market resilience and identify growth opportunities where AI complements human work.
Government Strategy
The government is developing several capabilities to manage AI risks. These include building internal evaluation capacity for public service deployment, expanding the AI Security Institute, advancing measurement science through the Centre for AI Measurement, and creating the AI Playbook for UK Government.
Steve Hasker, President and CEO of Thomson Reuters, said the partnership reflects the UK's position as a growing center for AI innovation. "What stood out to us was the open mindedness and truly collaborative approach to partnership and research we found at Imperial," he said.
The lab sits within Imperial's School of Convergence Science, part of the university's Science for Humanity strategy. The school focuses on four themes: Health and Technology, Human and Artificial Intelligence, Space Security and Telecoms, and Sustainability.
Learn more about Generative AI and LLM Courses or explore AI Research Courses to deepen your knowledge of frontier AI development.
Your membership also unlocks:
