DeepMind CEO Demis Hassabis says AGI may arrive by 2029, warns world is unprepared

Google DeepMind chief executive Demis Hassabis has issued one of his starkest warnings yet on the future of artificial intelligence, saying humanity may already be approaching the early stages of the “singularity” as progress towards artificial general intelligence accelerates faster than many expected.

Speaking after Google’s annual developer conference, Hassabis said recent breakthroughs in AI agents and autonomous systems have increased confidence across the industry that AGI, broadly described as AI capable of matching or surpassing human-level intelligence across tasks, could emerge before the end of this decade. While he still broadly points to 2030 as a likely milestone, he now sees 2029 as a realistic possibility and suggested developments could move even faster.

He argued that the rapid improvements seen in AI tools over the past year are not isolated advances, but signs that the industry may finally be on the right technical path towards far more capable systems.

AI agents are becoming a warning sign

According to Hassabis, the rise of AI agents capable of carrying out tasks independently should be treated as an early stress test for society. He described the current “agentic era” as a preview of how disruptive future AI systems could become once they gain stronger reasoning, planning and research abilities.

More from Tech

“We can see agents really happening now,” he said, pointing to how quickly such systems are becoming more useful in software development, automation and productivity tasks.

He added that the pace of improvement has already caught businesses and governments off guard. Recent examples involving advanced AI models from rival firms, including Anthropic, demonstrated how quickly new systems can surprise even industry insiders. Hassabis suggested such incidents should serve as a “warning shot” that society remains underprepared for what is coming next.

Although AI systems are not yet capable of fully improving themselves independently, Hassabis said early forms of what he described as “soft self-improvement” are already visible. Coding assistants and research agents are helping engineers work faster, indirectly accelerating AI development itself.

One of the industry’s biggest long-term concerns remains recursive self-improvement, where AI systems begin materially contributing to their own advancement. Hassabis acknowledged that leading laboratories are closely monitoring this possibility because of both its immense potential and its risks.

Calls for stronger safety measures grow louder

Hassabis also used the conversation to push for faster government action around AI safety and regulation. He argued that discussions about AGI and its broader economic consequences remain largely limited to technology circles, while policymakers and economists continue underestimating the speed of change.

“My economist friends are still not taking this seriously enough,” he said, stressing that the debate needs to move beyond Silicon Valley and into mainstream policy planning.

He welcomed early discussions around stricter AI oversight, including proposals that could require companies to conduct safety testing before releasing advanced models publicly. However, he believes current efforts are moving too slowly given the rate of progress.

Hassabis confirmed that discussions are taking place between major AI companies regarding future safety frameworks, though he declined to provide specific details. He said the next few years could prove critical in determining whether societies adapt responsibly to increasingly powerful AI systems or become trapped in cycles of excitement, fear and backlash.

For now, he believes the world still has a narrow window to prepare. But with AI capabilities improving at a relentless pace, Hassabis warned that window may not remain open for long.