Emerging Global Hub for AI Safety: Why It’s Not in Silicon Valley

In early 2023, Anthropic, an American AI company, unveiled its advanced model, Mythos, to select businesses, sparking immediate speculation about its cybersecurity threats and hacking functionalities. The AI Security Institute (AISI), based in London and backed by the British government, further assessed this model, shedding light on its capabilities. Learn more about the evaluation.

London, particularly the King’s Cross tech district, is quickly becoming a significant AI hub outside the US, with major players like Anthropic and OpenAI—both originally based in San Francisco—establishing key offices here. The UK’s focus diverges from that of other global cities, prioritizing AI safety and the understanding of emerging model risks over merely competing with the US and China for dominance in size and power. AISI serves as a prominent organization evaluating frontier models submitted voluntarily by firms, solidifying the UK’s reputation as a leader in global AI safety.

David Leslie, a researcher at Queen Mary University of London specializing in ethics and technology, emphasizes, “The UK has led the way in ethical considerations, responsible AI practices, and governance for quite some time.”

The AI safety landscape in London also features Apollo Research, which investigates how AI models can potentially “plan” against humans, along with nonprofit GovAI, which formerly operated as the Center for AI Governance. You can read about AI management through computing resources.

Academic institutions like University College London, the Alan Turing Institute, and the University of Oxford contribute to this ecosystem, nurturing AI research talent.

While the UK’s emphasis on AI safety has carved out a successful niche, some experts argue that AISI should not act as a substitute for regulatory bodies. There are concerns that an overly narrow understanding of AI safety risks may lead to insufficient attention on larger societal issues, such as transparency and the implications of widespread AI deployment.

London’s AI Landscape

Central to London’s AI ecosystem is DeepMind, an AI company founded in 2010, which Google acquired in 2014. Co-founder Demis Hassabis has assured that the company will continue its operations in London. Notably, several graduates have founded successful startups in the city, including Ineffable Intelligence, led by David Silver, who secured $1.1 billion in seed funding earlier this year.

According to Hassabis, being in the UK “affords a unique perspective” compared to the competition found elsewhere. While a “hype culture” is indeed emerging within the UK AI scene, Leslie notes that there remains a cautious, evidence-based methodology to evaluate scientific claims.

To attract foreign businesses, the British government positions the UK as more stable than the US and more conducive to innovation than the European Union. Earlier this year, in March, during an ethical dispute with the Pentagon, London’s mayor, Sadiq Khan, praised Anthropic and encouraged their expansion within the city—a move they made in April. Technology firms perceive the UK, with its relatively strict EU regulations, as aligning closely with their business interests.

Companies in London feel an abundance of intellectual talent and find a welcoming regulatory climate that facilitates growth, according to experts in the field.

Prioritizing Safety

On the forefront of assessing AI risks is AISI, initiated by former chancellor Rishi Sunak at the inaugural Global AI Safety Summit in Bletchley Park, Milton Keynes, UK, in 2023—marking the UK’s leadership in the global quest for AI safety.

Chief Technology Officer Jade Leong articulates that AISI emerged from “a strong belief that governments must grasp the intricacies of AI.” AISI conducts industry-wide evaluations to ensure models function as intended and assess their potential influence on public perception. Companies, including Google DeepMind, OpenAI, and Microsoft, have submitted their models for rigorous testing, including checks for capabilities such as biological weapon creation. AISI researchers aim to identify vulnerabilities in models to enhance their safety.