I — Our Goal
Preventing AI catastrophe.
Humanity faces multiple challenges, all of which suffer from coordination issues. We recognize the need to act at scale across countries — but we struggle to do so. Since we can't take on every challenge at once, we have chosen to start with the most urgent: existential risk from superintelligence.
To reach a good, humanist future — where society benefits from artificial intelligence without facing catastrophic risks — we need guardrails that put safety first: binding regulation, international agreements, and a public that understands what is being built in its name. Today none of these exist, and too few people know how fast AI systems are advancing.
Leading researchers warn this could happen sooner than almost anyone is prepared for.
A coordination crisis
Climate, pandemics, AI — the pattern repeats: everyone sees the problem, but no one can act on it together. Fixing that is the project.
A humanist outcome
A good future requires putting safety first — so society can benefit from AI without facing catastrophic risks.
II — Why AI is different
These models are
grown, not built
Today's frontier AI models are not engineered the way software has been for the last fifty years. The difference matters more with every jump in capability.
Traditional software · BUILT
Frontier AI · GROWN
Traditional software · Built
Specified by engineers, line-by-line.
Frontier AI · Grown
Emerges from training on vast datasets.
Traditional software · Built
Behavior is intended and predictable.
Frontier AI · Grown
What it can and can't do is only discovered after training — not decided in advance.
Traditional software · Built
Failures can be debugged and patched.
Frontier AI · Grown
Failures can be unexpected and unexplainable.
Traditional software · Built
Capabilities advance with deliberate releases.
Frontier AI · Grown
Capabilities can leap with each new training run.
What's so risky about chatbots?
Today's chatbots are the first products of a technology being scaled toward something far more powerful.
Scroll for the full spectrum →
A Compressed Timeline
Scroll for more →
From research curiosity to critical milestone — in under a decade.
2020
GPT-3
A surprising hint that scale alone unlocks new abilities.
2022
ChatGPT
100 million users in two months. The world starts talking to machines.
2023
The warning
CEOs of the leading AI companies sign a statement: AI poses extinction-level risk.
2023
Agentic systems
Models start using software tools and carrying out multi-step tasks on their own.
2024
Reasoning models
Frontier models begin outscoring human experts on PhD-level benchmarks.
2025
Deceptive scheming
Apollo and METR document models deceiving their evaluators.
2026
The buildout
Agents complete tasks that take human experts hours; $700B is committed to AI infrastructure in a single year.
Next
Critical milestone
Leading researchers expect little advance notice if control is lost.
Want the full story?
Follow the year-by-year record of how AI got this far in our AI Chronicle.
The Analogy
Just as an ant colony doesn't understand why its hill is destroyed when humans bulldoze it to build a road, we will likely not grasp the reasoning or actions of systems that are…
- ...autonomous Acting without direct human instruction.
- ...agentic Pursuing goals across long horizons.
- ...embodied Operating in the physical world.
- ...inter-connected Coordinating across systems and networks.
- ...superintelligent Surpassing human capability across domains.
Things will just happen to us that we may not understand, intend, or control. Since no AI company can predict or control emerging models, it's in everyone's interest that no company or country builds superintelligent AI.
III — Our Position
Humanity's future cannot be negotiated behind closed doors.
01 · Outreach
Bringing the conversation to the people.
We show up where people already are — community events, classrooms, local media — and make the case for taking AI risk seriously in plain language.
02 · Education
Building the on-ramps.
Reading paths, briefings, a glossary, and training that turn curiosity into real understanding.
03 · Advocacy
Fighting for reasonable regulation, written by humans.
Members contact lawmakers, attend hearings, and support comprehensive AI safety frameworks that prioritize human flourishing.
IV — Our Principles
A humanist approach to existential risk.
We believe a flourishing human future requires more than survival. It requires institutions, norms, and political incentives that hold advanced AI development to human goals and values.
- 01
Coordination at scale
Individual action matters, but collective coordination creates the leverage necessary to influence systems.
- 02
Participation that fits real life
People contribute at different levels. Participation has to meet them where they are, not demand uniformity.
- 03
Measurable progress
Real accountability requires tracking what actually changes — outcomes, not intentions.
- 04
Humanist values
Technology should serve human flourishing — survival is the floor, not the goal.
Are you ready to
carry the torch?
Members come from every discipline. The common thread is taking the work seriously.
Apply to join, contribute to a project, or start with reading and conversation. Every level of commitment makes a difference.