Artificial General Intelligence (AGI)

AGI usually means AI that can learn and perform a wide range of intellectual tasks at or above human level, rather than a single narrow task. There is no agreed definition. Google DeepMind proposes levels based on performance and generality, while others focus on economic value or on how efficiently a system acquires new skills.

Researchers use several benchmarks together because each saturates. Examples are ARC-AGI for novel abstract reasoning, Humanity's Last Exam and GPQA for expert knowledge, GAIA for tool-using assistants, SWE-bench for software tasks and METR's task time horizons for long autonomous work.

Opinions differ and depend on the definition. Some researchers see early general capabilities in frontier models, while others point to gaps in reliability, long-horizon autonomy, learning from few examples and physical-world competence. Treat claims with care and compare several benchmarks and independent evaluations.

Common directions are scaling pre-training, reasoning with more test-time compute, world models and embodied learning, agentic systems with tools and memory, and automating parts of research itself. Each has open problems.

Concerns include misuse, loss of control over autonomous systems, deceptive behaviour, concentration of power and societal disruption. Labs publish scaling and preparedness frameworks, and governments run evaluation institutes, but independent assessment is still developing.

There is no single AGI law. Relevant instruments include the EU AI Act, the NIST AI Risk Management Framework, the OECD AI Principles, national safety institutes such as the UK AI Security Institute, and voluntary lab commitments.

Links were checked in November 2026. Some sites block automated checks and may need a manual click. This page is a resource list, not advice or a forecast, and lab documents reflect the labs' own views.