The race toward Artificial General Intelligence (AGI) is intensifying, but Google DeepMind’s leadership is urging caution. In a recent January 2026 interview, Koray Kavukcuoglu, CTO of Google DeepMind and Google’s Chief AI Architect, made a striking admission: despite rapid advances in artificial intelligence, the industry still does not know how to build AGI.
“We do not have the recipe,” Kavukcuoglu said, emphasizing that AGI remains firmly in the realm of research rather than an engineering certainty. His comments come amid growing excitement and competition driven by OpenAI CEO Sam Altman, who has repeatedly stated that AGI is the thing he is “most excited about ever in life.”
AGI Is Still an Open Research Problem
Kavukcuoglu’s remarks reflect a grounded and transparent view of the current state of AI. While modern models can reason, code, and generate human-like responses, he stressed that no definitive technical blueprint exists for creating a system with true general intelligence comparable to humans.
Rather than chasing raw capability milestones, Google DeepMind is focusing on understanding how AI systems become genuinely useful. According to Kavukcuoglu, this means relying heavily on “user signals” feedback from Google’s billions of users across Search, Android, and productivity tools to guide how AI evolves in real-world contexts.
Safety First, Not as an Afterthought
A major pillar of Google DeepMind’s philosophy is safety by design. Kavukcuoglu argued that if AGI is ever achieved, it must be developed with security, alignment, and control embedded from the start. Retrofitting safety measures after deployment, he warned, is not a viable strategy for systems that could eventually surpass human-level reasoning.
This approach contrasts with concerns raised across the industry that competitive pressure could encourage companies to prioritize speed over safeguards.
Competitive Tension With OpenAI
The comments also highlight a growing philosophical divide between Google DeepMind and OpenAI. Sam Altman has been vocal about his enthusiasm for AGI and reportedly declared an internal “Code Red” at OpenAI following Google’s release of Gemini 3 in November 2025. Gemini 3’s advanced reasoning abilities and its capacity to generate interactive applications directly from search queries were widely seen as a major leap forward.
Analysts note that Google holds a significant infrastructure advantage in this race. Unlike OpenAI, which depends heavily on Microsoft and Nvidia, Google controls the entire AI stack from custom Tensor Processing Units (TPUs) to global data centers and massive consumer distribution platforms.
A Broader Industry Debate
The debate echoes views from other AI leaders, including Microsoft’s Head of AI Mustafa Suleyman, who has consistently advocated for strong human oversight and control over superintelligent systems.
As excitement around AGI grows, Kavukcuoglu’s message serves as a reality check: progress is real, but the destination remains uncertain. For now, the path to AGI is less about having the right recipe—and more about asking the right questions.
