Safe AI: British Universities Unite Their Forces

An Academic United Front Against AI Risks

The UK is speeding up. Major British universities have launched a collaborative research initiative aimed at ensuring the safe and ethical development of artificial intelligence, according to The Telegraph^[45]. This mobilization is not isolated: it is part of a national strategy built over several years, which has just reached a new level with the announcement of the BOLD laboratory, hosted by Oxford and funded with £60 million by UKRI and EPSRC^[1].

This laboratory will bring together researchers from Oxford, UCL, and Imperial College London around open learning and AI safety, with £2 million specifically dedicated to recruiting PhD students^[2]. Oxford thus positions itself as a central integrator between the academic world, industry, and the British AI ecosystem — an ambition clearly stated.

LASR: AI Security as a National Resilience Issue

Alongside BOLD, Oxford already leads the Laboratory for AI Security Research (LASR), an £8.22 million partnership involving the Alan Turing Institute and the British government^[3]. The stated goal: to strengthen the country's cyber resilience against malicious uses of AI. The Turing Institute will lead research commissioning on this subject through open calls and targeted partnerships with universities^[4].

This architecture is not anecdotal. It reflects a British desire to treat AI security as a sovereignty issue, on par with traditional cybersecurity. The country is betting on robust fundamental research rather than solely reactive regulatory measures.

The AI Giants Join the Dance

The private sector is not left behind. Google DeepMind announced an expanded partnership with the British AI Security Institute (AISI), through a memorandum of understanding focused on fundamental research in security, including model and data sharing^[5]. This approach echoes commitments made as early as 2023 by OpenAI, Google DeepMind, and Anthropic, which had pledged to provide priority access to their models to support research on evaluation and security, with £100 million allocated to the British AI task force^[11].

The AI Security Institute itself, formerly AI Safety Institute, was created in 2023 following the Bletchley Park summit, with the mission of providing governments with a scientific understanding of advanced AI risks^[12]. It has since published a research agenda focused on the most complex technical challenges in security, with the ambition of building a solid scientific foundation for policymakers^[10].

A Cooperation Beyond Borders

The UK is not going it alone. The UK AI Safety Institute and the French Inria have partnered to advance AI safety research as part of an international scientific cooperation^[6]. This Franco-British dynamic is also embodied in the Entente CordIAle, a partnership uniting the Saclay Cluster, Oxford, and Cambridge around ethical and sovereign AI^[7].

• Structuring long-term cooperation in research, training, and innovation^[8]
• Five strategic axes including academic mobility and collaborative projects^[8]
• A £31 million Responsible AI UK (RAI UK) consortium, led by Southampton, funded by UKRI^[9]

This latest consortium explicitly aims to create a British and international research ecosystem on responsible and trustworthy AI^[9], proving that the ambition far exceeds the national academic framework alone.

Accumulating Public Funding

On the budgetary front, the British government has committed, through its AI Safety Institute, an initial fund of £4 million to finance research on risks related to AI technologies, with an aim to scale up to £8.5 million^[13]. These sums add to the £60 million for the BOLD laboratory and the £8.22 million for LASR, outlining a substantial cumulative financial effort for a country that intends to remain competitive against the United States and China in this specific scientific field.

Yet, the multiplication of structures — BOLD, LASR, AISI, RAI UK, Entente CordIAle — raises a legitimate question: how do these initiatives, carried by different institutions with distinct funding, concretely articulate on the ground? The risk of duplication of efforts is not negligible, even if each structure claims a specific scope, between cybersecurity, applied ethics, and international governance.

What It Changes Concretely

For researchers, this buzz translates into a multiplication of opportunities: doctoral scholarships, fellowships like the Cambridge ERA:AI, or the MARS V program organized in Oxford in July 2026. For tech companies, it's a clear signal: the UK wants to become an essential hub for testing and validating AI systems' safety before large-scale deployment.

For British citizens, the issue is more diffuse but just as real: ensuring that AI systems gradually infusing the economy and public services are subject to rigorous scientific control, and not left to the sole self-regulation of the companies developing them. It remains to be seen whether this dense institutional architecture will produce tangible results, or if it will remain a stack of structures with overlapping objectives.