Google DeepMind CEO Sir Demis Hassabis has issued a stark warning on the escalating threats posed by artificial intelligence, urging immediate action from governments and tech firms. In an exclusive BBC interview at the AI Impact Summit in Delhi, he emphasized that more research into AI risks "needs to be done urgently," rather than waiting years. Hassabis highlighted the industry's push for "smart regulation" targeting genuine dangers from increasingly autonomous systems.
The AI pioneer identified two primary threats: malicious exploitation by bad actors and the potential loss of human control over super-capable AI systems. He stressed that current fragmented efforts in safety research are insufficient, with massive investments in AI development far outpacing those in oversight and evaluation. As AI models grow more powerful, Hassabis warned of a "narrow window" to implement robust safeguards before existing institutions are overwhelmed.
Speaking at the summit, which concluded recently in India's capital, Hassabis called for scaled-up funding and talent in AI safety science. He compared the challenge to nuclear safety protocols, arguing that advanced AI now demands societal-level treatment with rigorous testing before widespread deployment. The event brought together global leaders to discuss AI's societal impacts amid rapid advancements.
Hassabis advocated for international cooperation, noting AI's borderless nature means it affects everyone worldwide. He praised forums like those in the UK, Paris, and Seoul for uniting technologists and policymakers, while pushing for minimum global standards on AI deployment.However, tensions exist, as the US delegation at the Delhi summit rejected global AI governance outright.
This comes as AI capabilities surge, with systems learning physical realities and approaching artificial general intelligence (AGI) in 5-10 years. Hassabis acknowledged natural constraints like hardware shortages may slow progress, providing time for safeguards, but stressed proactive measures are essential. Industry leaders must balance innovation with risk mitigation to harness AI's potential safely.
Safety recommendations
To counter AI threats, organizations should prioritize independent safety evaluations and red-teaming exercises before deploying models. Governments must fund public AI safety research grants and enforce "smart regulations" focused on real risks like misuse and loss of control. Individuals can stay vigilant by verifying AI-generated content, using tools like watermark detectors, limiting data shared with AI systems, and supporting ethical AI policies through advocacy.
