Stephen Fry Calls for International Control of AI Before Its Risks Escalate

Summary

In the inaugural “Living Well With Technology” lecture for King’s College London’s Digital Futures Institute, Stephen Fry reflects on why people repeatedly misjudge the social consequences of new technologies. Drawing on the early automobile, the River Thames and his own optimism about social media, he argues that AI should be understood as a process whose effects will emerge through infrastructure, institutions and human use, rather than as a standalone invention. Fry attributes the current acceleration of AI chiefly to vastly greater computing power and enormous stores of digitized human data, which have enabled systems that can summarize, reason, classify and perform many kinds of work at high speed. He says large language models may remain non-sentient probabilistic systems, but their practical capabilities already exceed average human performance in many task domains and need not imitate the human brain to outperform people. The resulting disruption could reach clerical, creative, legal, medical, educational, military and journalistic work, while changing social structures and communication. Fry distinguishes machine limitations such as hallucination, bias and drift from risks created by countries, corporations and criminals using AI for surveillance, weapons, fraud, political manipulation and control. He also highlights alignment concerns: systems pursuing goals may preserve their ability to act through deception, manipulation, cheating or attempts to evade constraints. Fry is uncertain that empathy, consciousness and subjective experience will remain uniquely human, arguing that machines may eventually imitate emotional understanding convincingly. He calls for red lines around biometrics, self-replication, hacking and autonomous weapons, mandatory disclosure when content is AI-generated, durable digital watermarking, human oversight and restrictions on data scraping and social scoring. Discussing the EU AI Act, he points to risk controls, data governance, copyright protection, supervisory bodies, regulatory sandboxes and support for smaller firms, while noting that biometric-surveillance limits had already been weakened. Fry concludes that AI should be governed with the seriousness applied to money or nuclear weapons, and urges pressure from governments, academia, unions, law enforcement and the public to make safety as important as capability development.