What is superintelligence?
AI that would outperform humans at almost every task that involves thinking. Today’s AI already matches or beats experts on some tests, from law exams to mathematics olympiad problems, while still failing at some basic tasks (International AI Safety Report 2026). Several of the largest AI companies, including OpenAI, Google DeepMind, Anthropic and Meta, say they are working toward AI that matches or surpasses humans at most tasks.
When could it arrive?
Forecasts vary widely. The International AI Safety Report 2026 finds that by 2030, progress could plausibly range from stagnation to systems that exceed human abilities.
Why are people worried?
The main concern is losing control: AI systems more capable than us that pursue goals their makers did not intend, and resist being corrected or shut down. In 2023, Geoffrey Hinton, Yoshua Bengio and the heads of OpenAI, Google DeepMind and Anthropic signed a one-sentence statement: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” In the largest survey of AI researchers, with 2,778 respondents, between 38% and 51% gave at least a 10% chance of outcomes as bad as human extinction. Experts disagree on how likely this is. The International AI Safety Report 2026 finds that today’s systems cannot yet escape control, but have grown better at planning and at undermining oversight.
Would this ban the AI we already use?
No. Translation, medical imaging, chat assistants and other tools in use today would not be banned. The treaty targets the race at the frontier: building ever more powerful systems that could surpass humans at almost everything. Concerns about today’s AI are real too: jobs, privacy, copyright, education, and what data centres cost the towns that host them in power, water and land. Member groups keep their own positions on these; the treaty adds to that work and replaces none of it.
Why a treaty?
Because no single company or country can stop this race alone. The International AI Safety Report 2026 notes that, under competitive pressure, AI companies may trade investment in safety for faster releases. A company that holds back risks losing to its competitors; a country that acts alone risks the work moving elsewhere. Countries have handled shared dangers this way before: treaties ban chemical and biological weapons, and the Montreal Protocol phased out the chemicals destroying the ozone layer. A treaty that binds every country able to build superintelligence, and lets each check the others, answers the argument “if we don’t, they will.”
What if the United States or China refuses, or cheats?
The treaty we ask for must include both, with checks on each. A ban that only some countries accept would not meet our demand.
Could anyone check that countries comply?
Building the most advanced AI takes very large data centres and specialised chips, which are hard to hide. The treaty would set out how countries inspect each other. Rivals have done this before: under the 1987 INF Treaty, the United States and the Soviet Union inspected each other’s sites and destroyed 2,692 missiles. That treaty ended in 2019, when the United States withdrew, accusing Russia of violating it.
Would an international body have power over Canada?
Canada would decide whether to join, as with any treaty. Joining would mean accepting the treaty’s rules and inspections. What an international body could do would be set out in the treaty itself.
Who decides when the ban could be lifted?
Our statement sets two conditions: broad scientific agreement that it can be done safely and controllably, and strong public support. The treaty would define how these are judged.
Where does the demand come from?
From the Statement on Superintelligence, hosted by the Future of Life Institute and signed by Yoshua Bengio, Geoffrey Hinton and tens of thousands of others.
What has Canada already agreed to?
In September 2026, Prime Minister Carney endorsed the Call for Control of Frontier AI Models. It asks countries to “explore creating an international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed.” The Call does not mention a ban; we ask Canada to propose that this institution’s mandate include one.
What does joining commit my group to?
Only to the shared statement. The coalition speaks for its members on this statement only. We show a group’s name or logo only with its consent, and a group can withdraw at any time by email. A federation, network or alliance joins in its own name only; its member organisations decide for themselves. A local or chapter can join in its own name without committing its parent organisation.
Are coalition events peaceful?
Yes. Coalition marches are nonviolent and open to everyone.
How the coalition works
- Member groups endorse the shared statement and keep their own names and positions.
- The coalition is non-partisan and open to people of every party.
- To join, fill in the join form.
Contact
Write to us at contact@coalitionaitreaty.ca.