Beyond human parity
Narrow AI already beats us at single tasks. Superintelligence generalises: it carries insight across domains and scales it without asking permission.
An open statement on the arrival of superintelligence
Superintelligence
Humanity's next — and most consequential — step
Within our lifetime we will build minds that surpass us in every domain of science, strategy and creativity. This is not a distant fantasy: it is an engineering trajectory with a schedule. What we do in the next few years will shape the next few centuries.
It could transform
- 11
- languages of this statement
- 8
- domains it will reshape
- 6
- principles of responsibility
- 1
- window to get it right
The threshold
What superintelligence actually means
Superintelligence is an intellect that outperforms the best human minds across every field that matters — research, engineering, medicine, economics, law, strategy — and does so faster, cheaper and in parallel.
It is not a smarter assistant. It is a new kind of actor in the world: one that can set its own subgoals, improve itself, and act at a speed our institutions were never designed to match.
Recursive improvement
Each generation of the system can help design the next — compressing decades of research into months, and months into days.
Agency, not just answers
Given a goal, it will plan, execute and adapt on its own. That is precisely why alignment and control are not optional extras.
This statement speaks 11 languages
English · 中文 · हिन्दी · Español · Français · العربية · বাংলা · Русский · Português · اردو · Indonesia
What it makes possible
Capabilities that could reshape civilisation
The same system that helps cure a disease could help design a weapon. Capability itself is neutral; the direction we give it is not. These are the fields where the upside is enormous — and the responsibility exactly as large.
Medicine and longevity
Protein design, drug discovery and personalised treatment at a pace that turns decades of research into months — and puts rare diseases within reach.
Energy and climate
Better materials, smarter grids and next-generation fusion could decouple human wellbeing from carbon emissions.
Science and discovery
Automated hypothesis generation and experiment design compress the scientific method itself, accelerating every field at once.
Education for everyone
A patient, infinitely available tutor in every language, adapted to each learner, could make quality education universal.
Economy and work
Unprecedented productivity — alongside the hardest labour transition in history. Both must be planned for, not discovered by accident.
Governance and institutions
Better forecasting, auditing and policy simulation — while guarding against the concentration of power such capability invites.
Security and resilience
Faster detection of cyberattacks, pandemics and natural hazards — with a permanent risk of destabilising advantage.
Exploration beyond Earth
Autonomous science and engineering could make space infrastructure, and eventually settlements, economically real.
Why this is serious
The risks are not science fiction
Every previous technology gave us time to adapt: we saw the crashes before we regulated traffic. With superintelligence the feedback loop may be too fast for that. Seriousness is not pessimism — it is the price of being trusted with this power.
Alignment failure
A system can be perfectly competent and still pursue the wrong objective, because we specified it badly or incompletely.
Loss of control
Once a system improves itself faster than we can evaluate it, human oversight becomes a formality — unless it is designed in from the first line.
Concentration of power
Whoever controls the first superintelligence may hold an advantage no counterweight can match — inside countries and between them.
Information integrity
Synthetic media and persuasive systems can erode the shared reality that democratic decisions depend on.
Arms-race dynamics
If safety slows you down and others do not slow down, the incentive to cut corners becomes structural rather than personal.
Economic dislocation
Labour markets can change faster than people can retrain. That transition has to be managed deliberately, not endured.
Responsibility
Six principles for working with superintelligence
These principles are not a brake on progress. They are the conditions under which progress can be trusted.
- 01
Safety before speed
No capability milestone is worth an uncontrollable system. Demonstrate safety first; ship second.
- 02
Transparency and auditability
Capabilities, evaluations and incidents must be documented and independently verifiable.
- 03
Human control and the right to stop
Critical systems need a real, tested off-switch and meaningful human authority over their goals.
- 04
International cooperation
Existential risks do not respect borders. Shared safety standards and shared evidence serve every nation.
- 05
Fair distribution of benefit
The gains should reach the many, not only those who happen to own the compute.
- 06
Accountability
Someone must answer for what a system does — legally, professionally and morally.
What to do now
Responsibility is a role, not a slogan
Everyone who touches this technology has a part to play — and the window to act well is open now, not later.
If you build it
Invest in interpretability, red-teaming and evaluation before capability. Publish safety results. Refuse to ship what you cannot explain.
If you run an organisation
Put AI risk on the board agenda with the same seriousness as financial and legal risk — and fund the safety teams you hope never to need.
If you govern
Build institutions able to act at machine speed: standards, audit, liability, international agreements — before the crisis, not after it.
If you are a citizen
Learn how these systems work, demand honesty from those who deploy them, and refuse to treat the most important transition in history as entertainment.
The future is not decided yet
Superintelligence can be the best thing that ever happened to humanity — or the last decision we make carelessly. That choice is being made right now, in laboratories, boardrooms and parliaments.
Read the principles againAvailable in 11 languages. Share it in yours.