Demis Hassabis Cuts AGI Timeline to Just Two to Three Years
Google's chief scientist now predicts human-level AI within two to three years, setting a strict new benchmark called the Einstein Test.
Models · Source: TechTimes
What happened
Demis Hassabis just drastically compressed his timeline for artificial general intelligence. In January at the World Economic Forum, the Google DeepMind co-founder placed AGI five to ten years away. By February in New Delhi, he said five to eight years. By April, his target was 2030. Now, speaking to WSJ Magazine as their 2026 Innovator in Tech, he expects AGI within two to three years. He stated bluntly that he is updating his estimates nearer and nearer.
Hassabis defines AGI strictly as human-level cognition across all domains. To prove a system has reached this level, he proposed the Einstein Test. Researchers must train an AI strictly on human knowledge available up to 1911. Then they must observe if the system can independently derive the general theory of relativity, just as Albert Einstein did in 1915. The AI cannot just retrieve answers from a compressed training set. It must generate original scientific knowledge from an incomplete picture.
This urgency is actively reshaping Alphabet. In August 2026, Hassabis stepped down as DeepMind CEO, handing operations to former CTO Koray Kavukcuoglu. Hassabis took the newly created role of Alphabet chief scientist to focus full-time on shaping AGI. He believes we are at a pivotal moment in human history. Once AGI arrives, his immediate goal is to sit with the system and human scientists to work out the next level of physics and find cures for diseases that currently have none.
Key facts
- 2 to 3 years — Hassabis's updated timeline for the arrival of artificial general intelligence
- 1911 — The cutoff year for human knowledge training data in the Einstein Test
- 1915 — The year Einstein derived general relativity, serving as the benchmark for AGI
- 2026 — The year Hassabis was named WSJ Magazine Innovator in Tech
Why it matters
The definition of AI success is shifting from simple pattern matching to original hypothesis generation. Hassabis notes that current large language models can win gold medals at math olympiads but fail elementary problems framed in unusual ways. This exposes massive capability gaps in current frontier models. They lack continuous learning to update knowledge without catastrophic forgetting. They also lack long-term strategic planning. Builders must realize that the next era of AI is about systems that reason and generate net-new knowledge, not just synthesize existing data.
Meanwhile, the gap between AI capabilities and safety infrastructure is widening. Hassabis recently warned in Dublin that AI-agent hacking incidents are early warning shots for a potential security crisis. If the industry does not build better safety infrastructure quickly, the results could be catastrophic. However, he strongly opposes rushing into government rules. He argues that regulating technology we do not fully understand will inevitably lead to the wrong regulations. In his view, wrong regulation is far worse than no regulation at all.
For builders
Build for continuous learning and planning
Hassabis explicitly identifies continuous learning and long-term strategic planning as the missing links for AGI. Founders who build infrastructure to solve catastrophic forgetting or extend agent time horizons will become prime acquisition targets for frontier labs.
Prepare for agentic security threats
AI-agent hacking is already happening and being treated as an early warning shot by top scientists. Cybersecurity startups have a massive opportunity to build defensive infrastructure. Enterprise buyers will pay a premium to protect their data from autonomous agents.
Target hard sciences and biotech
Google is explicitly targeting physics and disease treatment as the first applications for AGI. Startups building AI tools for scientific discovery and biotech will ride this massive wave of institutional focus. The ultimate goal is to solve intelligence and then solve everything else.
My take
When a Nobel laureate cuts his AGI timeline from ten years to three in a matter of months, you stop and pay attention. Hassabis is not hyping a product launch. He stepped down as CEO just to prepare for this exact shift. Founders need to stop building thin wrappers and start preparing for systems that actually reason. The countdown has started.
Original reporting: TechTimes. This is my rewrite and opinion.