From Famagusta roots to AI’s top table, Hassabis calls for a global watchdog

Sir Demis Hassabis, the son of a Greek Cypriot father from Famagusta, is using his move away from the day-to-day running of Google DeepMind to press for independent oversight of the world’s most powerful artificial intelligence models. The Nobel Prize-winning scientist discussed the proposal with leaders of other AI laboratories, government officials and European policymakers in the weeks before relinquishing his chief executive role, according to people familiar with the talks. In talks with peers, Hassabis compared the body’s eventual international role with that of the International Atomic Energy Agency, the UN-linked watchdog established to promote cooperation and safeguards in the nuclear field. His aim is to establish an industry-funded but independently governed organisation that would test frontier AI models before they are released, draw up common safety standards and intervene if systems displayed dangerous capabilities. The proposal brings together the two strands that have defined Hassabis’ career. One is the race to build increasingly intelligent machines. The other is his long-standing concern over what happens if their capabilities advance faster than society’s ability to control them. For Cyprus, his rise carries a particularly close connection. Hassabis was born in London in 1976 to a Greek Cypriot father, Costas, whose family came from Famagusta, and a Chinese Singaporean mother, Angela. Hassabis has also spoken directly about that connection. “It makes me happy that Cypriots are proud of me,” he said. “I have a Greek way of thinking,” Hassabis said, adding that what Hellenism has contributed to the Western world since the classical period “is a big part of who I am”. Raised in north London, he showed an early fascination with games and strategy. He learnt chess at four and became one of the world’s leading junior players, at one stage ranking second globally in his age group. His chess winnings paid for his first computer, a ZX Spectrum, when he was eight. With neither parent working in technology, he taught himself to programme from books. “Because I started playing chess when I was four, it really was very formative in how I think,” Hassabis said in an earlier interview, linking the game to his approach to strategy and pressure. By 17, he had helped design the hit computer game Theme Park. He later studied computer science at Cambridge, founded his own video game studio and eventually returned to academia, completing a doctorate in cognitive neuroscience at University College London. That combination of games, computer science and the study of the human brain shaped the company he would create. Hassabis co-founded DeepMind in London in 2010 with Shane Legg and Mustafa Suleyman. Its mission was deliberately ambitious, to “solve intelligence” and then use it to solve other problems. Google acquired the company four years later for a reported £400 million, giving its researchers access to the computing power needed to pursue far larger experiments. DeepMind first captured global attention with AlphaGo, the system that defeated champion player Lee Sedol at Go in 2016. The victory was significant because the ancient board game had long been considered too complex and intuitive for a computer to master at the highest level. However, Hassabis repeatedly stressed that games were only a testing ground. The real objective was to apply the same learning techniques to science, medicine and industry. That ambition produced AlphaFold, which transformed scientists’ ability to predict the three-dimensional structures of proteins. Understanding those structures can help researchers investigate diseases, identify drug targets and design new treatments. The achievement earned Hassabis and DeepMind researcher John Jumper half of the 2024 Nobel Prize in Chemistry, with the other half awarded to David Baker. AlphaFold has since been used to predict more than 200 million protein structures, making the resulting database a widely used tool in biological research. Now 50, Hassabis is moving into another phase of his career. Under Google’s leadership overhaul, he has become chair of Google DeepMind and chief scientist of parent company Alphabet, while longtime deputy Koray Kavukcuoglu has taken control of the laboratory’s daily operations. The reshuffle comes as Google faces intense competition from OpenAI and Anthropic, while pressure grows to turn costly AI research into commercial products. Alphabet shares fell 4 per cent after the changes were announced. For Hassabis, however, the move creates more time to concentrate on scientific discovery, the development of artificial general intelligence and the risks surrounding it. He believes AGI, generally understood as a system able to perform the full range of human cognitive tasks, may be only a few years away. In his proposed framework, he warned that laboratories were caught in an intense commercial and geopolitical contest while advances were outpacing their understanding of the technology. Under his plan, developers would initially submit their most capable models for independent examination up to 30 days before release. Tests would look for dangerous abilities involving cyberattacks, biological threats, autonomous behaviour and deception. Participation could begin voluntarily, allowing experts to develop reliable testing methods before any stronger rules were considered. The longer-term ambition is international, with shared standards intended to prevent companies or countries from gaining an advantage simply by taking greater risks. Existing organisations already carry out parts of that work. Britain’s AI Security Institute tests advanced systems, while the EU’s AI Act gives the European AI Office powers to examine general-purpose models, demand corrective action and impose fines when they pose systemic risks. Cyprus has its own role in that system. The island’s Communications Commissioner has been designated as the national notifying authority, market surveillance authority and single point of contact for the AI Act. Since August 2, the European AI Office and national authorities have begun enforcing the rules across the bloc, meaning frontier models offered in Cyprus fall within the same European oversight framework. Still, none fully matches what Hassabis is proposing, namely an independent technical referee able to inspect the leading models before the public encounters them and, eventually, stop one considered too dangerous to release. The challenge will be persuading rival laboratories to disclose closely guarded technology while they are competing for users, investment and talent. An organisation funded by the same companies it examines would also need strong safeguards to show that its independence was more than nominal.