London Emerges as a Global Hub for AI Safety

While the United States and China race to build ever-larger models, Britain is playing a different card: London is becoming the place where AI companies test their systems for risk before they reach the rest of the world. A thickening cluster in the King's Cross district and the work of the government's AI Security Institute (AISI) are turning the UK capital, rather than Silicon Valley, into the reference point for AI safety.

The King's Cross cluster

King's Cross, once known mainly for its railway station, has turned in recent years into a dense cluster of artificial intelligence companies. Google DeepMind, Isomorphic Labs, Synthesia and Wayve already operate side by side there, and they are now being joined by the European offices of Anthropic and OpenAI. Within a few blocks sit the headquarters of AstraZeneca and GSK, which makes collaboration on AI applications in medicine easier.

Proximity to University College London, where DeepMind chief Demis Hassabis studied, as well as Imperial College and the Alan Turing Institute, gives companies access to fresh graduates and researchers without having to relocate them from the US. Google is investing 3 billion pounds in the same district to build a 330-meter office tower, expected to open in late 2026.

Why Anthropic is betting on London

Anthropic's decision to quadruple its London team is not purely about proximity to talent. The company says it wants to deepen its cooperation with the AI Security Institute, one of the first bodies in the world to systematically test models for risks before they reach widespread use. Google DeepMind, OpenAI and Microsoft also submit their models for independent AISI evaluations.

Europe's largest businesses and fastest-growing startups are choosing Claude, and we're scaling to match - Pip White, Anthropic's head of EMEA North

Anthropic's earlier dispute with the Pentagon, in which it refused to let its models be used for mass surveillance and autonomous weapons systems, further drew the attention of British authorities to the company as a partner more open to independent oversight than some of its US competitors.

What the Claude Mythos Preview evaluation showed

The freshest evidence of the London hub's significance is the public evaluation of Claude Mythos Preview, the model Anthropic announced on April 7. AISI built a 32-step simulated attack on a corporate network, internally named The Last Ones. The model completed an average of 22 of the 32 steps, while the best previously tested model, Claude Opus 4.6, averaged 16.

Multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities autonomously - tasks that would take human professionals days of work - from the AI Security Institute's evaluation

The institute notes that the model failed part of the simulation involving operational technology, dubbed Cooling Tower, showing that AI's autonomous offensive capabilities still have limits, even as those limits shift quickly with each new generation of models.

A network of AI safety institutes

AISI does not operate in isolation. International leaders have agreed to form a network of AI safety institutes spanning the United Kingdom, the United States, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada and the European Union. The model developed in London, built on independent pre- and post-release testing, is thus becoming a reference point for regulators worldwide.

For Polish readers, this means that the safety standards for models used by Polish companies or public administration are increasingly being shaped in London rather than only in Brussels or Washington. Findings from evaluations like the one on Claude Mythos Preview later feed into guidelines that also underpin EU requirements for providers of high-risk models.

The growing concentration of AI labs and independent oversight bodies in one place also raises questions about who stands to profit financially from this growth. Of the later funding rounds for companies such as Isomorphic Labs, which raised 1.57 billion pounds, only about 10 percent of the capital came from British investors, with most flowing in from the US and global funds.