What kind of claim is this?
A definition, a documented position, a personal forecast or a hypothetical scenario?
Try this: Point to the words that tell you. Then check the original source.
100 BUTTONS / A THOUGHT EXPERIMENT
If one button could end humanity,
would you press any?
This board has ten.
A hypothetical choice. Your clicks here cause no real-world harm. Ten buttons open overlapping scenarios, not ten measured or independent extinction risks.
Why ten? The supplied screenshot quotes Evan Hubinger’s personal estimate of >10% within the next decade for AI killing all humans. Exactly 10/100 is our illustration, not a measured probability, consensus or Anthropic forecast. Read the context ↓
THE ESTIMATE BEHIND THE BOARD
EXPLORE THE CHOICES
Who gets to decide
what we risk?
Meet the people. Read the evidence. Ask better questions.
The real-world “buttons” are decisions about research, deployment and governance. The profiles explain documented roles and public positions, including arguments for benefits and safety. They do not accuse anyone of knowingly causing extinction.
Read leaders’ statements and safety commitments →Choose any number below. Read a short card. Answer an optional question for learning points, or keep exploring. No timer. No points for taking a risk.
10/100
Hypothetical scenarios.
The other 90 buttons revisit 40 ideas and public profiles. Numbers identify buttons; they are not clues.
EXPLORED000 / 100 EXPLORED
SCENARIOS00 / 10
Optional questions earn 10 points per new topic. You can always try again.
Unopened Learning Scenario
Pick any number. Numbers are labels, not Minesweeper clues. Ten buttons reveal hypothetical scenarios. How the board works.
All 100 explored. The next step is outside the board. Turn curiosity into a better question →
WHY TEN?
One estimate.
A reason to ask questions.
The embedded Evan Hubinger post in this supplied Ryan Orhan screenshot inspired the ten danger buttons. His personal estimate is greater than 10% within the next decade for AI killing all humans.
Our board uses exactly 10 out of 100 as a simplified design choice. It is not an exact restatement of his figure, a measured probability or scientific consensus. This is not Anthropic’s official forecast.
The screenshot preserves the commentator’s reaction and embedded quote. It documents displayed wording; we could not authenticate the original X post directly.
Original post link (not directly verified) ↗
Readable excerpt from the embedded quote: “I personally think it is >10% within the next decade.” Hubinger also says he believes Anthropic is trying its best but does not yet have a plan to solve alignment for superintelligence.
01 / UNDERSTAND THE PREMISE
A thought experiment.
Not a forecast.
Artificial superintelligence (ASI) means a hypothetical AI with abilities far beyond humans across a broad range of cognitive tasks. Today’s models are not established examples of that kind of system.
In September 2026, Ars Technica reported that Anthropic researcher Evan Hubinger personally put the chance of AI killing all humans at more than 10% within the next decade. This is one person’s judgement, not Anthropic’s official forecast. We could not directly verify the original X post; read the reporting ↗.
We selected exactly ten danger buttons as a visual reference point, not a precise reproduction of “more than 10%”. Experts disagree about the likelihood and severity of losing control of future AI. Uncertainty is a reason to examine the evidence, not proof that disaster is either inevitable or impossible.
What the board can and cannot tell you
Ten positions are randomly assigned when a board starts. Initially, a uniformly random choice has a 10-in-100 chance of revealing a scenario. As buttons open, the composition of the remaining board changes. Opening every button necessarily reveals every scenario. None of this calculates humanity’s cumulative risk.
The ten scenarios are overlapping educational sketches, not ten independent, proven extinction pathways. Some describe large-scale harm rather than extinction. The 90 other buttons revisit a library of 40 concepts and public profiles. Repetition is intentional.
EXPLORE TOGETHER
Make it a conversation.
You do not need to agree. You do need to ask good questions.
A suggested group format
- 01Set the scene
5 min · Read Why ten? together. Separate hypothetical ASI from today’s models.
- 02Explore a few buttons
10–15 min · Take turns choosing. Read the source, discuss the card and try an optional question.
- 03Compare your views
5 min · Choose one prompt below. Give everyone space to disagree and explain why.
What could we gain?
What could we lose?
Put a possible benefit beside a possible harm. Who experiences each?
Try this: Name a safeguard and explain what it would need to demonstrate.
Who gets a say?
Who builds, funds, deploys or governs this technology? Who carries the risk?
Try this: Use your own region. What practical routes exist to question a decision?
What would change your mind?
Choose a finding that would make you more concerned, or more reassured.
Try this: Explain how you would tell a strong result from a persuasive promise.
Before the session: display, sound and points
Make it readable. Open Options & group sessions beneath the board and turn on Presenter mode for larger reveal text. The discussion cards also enlarge. Sound stays off unless you choose it; agree with the group first.
Keep it low-pressure. Questions are optional. There is no timer, speed bonus, penalty or public ranking. Points record practice answers, not qualifications or agreement with a forecast.
Start fresh for a new group. Reset board clears this browser’s points and reshuffles. Personal devices keep separate progress. Remembering points between visits is optional.
Make it local. Invite examples from the places represented in your group. Political institutions and ways to participate differ around the world.
THE FUTURE IS SHARED
Start where you are.
Useful questions travel across borders. The right way to act depends on your community, institutions and circumstances.
Check, then share.
Read the original source, date, assumptions and uncertainty. Discuss what you learn with your community, class or colleagues. A striking claim is a reason to investigate.
Read the international evidence ↗Ask who is accountable.
Where available and appropriate, use local consultations, representatives, regulators or professional and community organisations to ask about AI testing, oversight and routes to challenge harm. Political rights and public processes differ.
Explore UNESCO’s shared principles ↗Support careful work.
Ask the organisations you use about safeguards and evidence. Follow independent research, share accessible learning, or contribute relevant skills to evidence-led safety work. Compare possible benefits, harms and who bears them.
Use the OECD principles as questions ↗Optional regional starting points
These resources cover different places and purposes. They are not equivalent laws, universal complaint routes or proof that safeguards work. Check current local procedures before acting.
- Africa: African Union Continental AI Strategy, a regional policy starting point.
- European Union: European AI Office, official information about EU AI governance.
- China and the United States: Compare named institutions and dated policies, including cooperation.
- United Kingdom only: Find your MP, if that parliamentary route is relevant to you.
03 / READ BEYOND THE BOARD
Curiosity needs evidence.
Primary research, company documents and clearly labelled reporting. Sources checked 13 September 2026. Roles and policies can change.
IBM: Artificial superintelligence ↗Technical explainer · definition used; older capability commentary not relied onIBM: AI agents ↗Technical explainer · agent and model distinctionConcrete Problems in AI Safety ↗Primary research · Amodei et al. · 2016Thousands of AI Authors on the Future of AI ↗Primary survey research · differing questions and forecasts · 2024Dario Amodei: We Must Pace the Frontier ↗Personal essay / stated company commitment · September 2026International AI Safety Report 2026 ↗Independent scientific synthesis · February 2026An Overview of Catastrophic AI Risks ↗Research paper · Hendrycks, Mazeika & Woodside · 2023Anthropic Responsible Scaling Policy v3 ↗Company policy · February 2026Statement on AI Extinction Risk ↗Advocacy statement and signatories · 2023Anthropic leadership ↗Company source · roles checked September 2026OpenAI: Our structure ↗Company governance · checked September 2026Google: The next chapter of our AI momentum ↗Leadership announcement · August 2026The reported Hubinger estimate ↗Secondary reporting · 9 September 2026 · original X post not directly verifiedDeepSeek-R1 发布，性能对标 OpenAI o1 正式版 — R1 release ↗DeepSeek · Chinese company announcement · 20 January 2025Qwen3：思深，行速 — Qwen3 release ↗Qwen team / Alibaba Cloud · Chinese company post · April 2025 releaseQwen3 Technical Report ↗Qwen team · Primary technical report · May 2025《人工智能安全治理框架》2.0版发布 — AI Safety Governance Framework 2.0 ↗Cyberspace Administration of China · Chinese official release · 15 September 2025全球人工智能治理倡议 — Global AI Governance Initiative ↗Chinese Ministry of Foreign Affairs · Chinese government position · 20 October 2023 publicationCracking the code: How Zhejiang University shapes the “Six Little Dragons” ↗Zhejiang University · Institutional profile · 4 March 2025 displayed date唐杰：坚定不移探索通用人工智能发展的中国道路 — Tang Jie on a Chinese path to AGI ↗Tang Jie · Signed essay hosted by Tsinghua · 12 August 2026 page; original 10 August照见未来，就是最好的信心答卷 — University-linked innovation ↗Tsinghua-hosted China Education Daily report · March 2026研究院介绍 — Institute for AI International Governance ↗Tsinghua I-AIIG · Chinese institutional description · undated; checked 13 September 2026李强出席2025世界人工智能大会暨人工智能全球治理高级别会议开幕式并致辞 — Li Qiang’s 2025 AI address ↗Chinese Ministry of Foreign Affairs · Official Chinese account · 26 July 2025李强会见美中贸易全国委员会董事会代表团 — Current role check ↗Chinese Ministry of Foreign Affairs · Official Chinese account · 1 September 2026BIS: Semiconductor export licence policy for China ↗US Bureau of Industry and Security · Official announcement · 13 January 2026America’s AI Action Plan ↗White House · US government policy plan · July 2025The Bletchley Declaration ↗UK government publication · Joint international declaration · 1 November 2023UNESCO Recommendation on the Ethics of AI ↗Global intergovernmental recommendation · adopted 2021; implementation variesOECD AI Principles ↗Intergovernmental principles · updated May 2024; not a universal lawAfrican Union Continental AI Strategy ↗Africa · Regional strategy · published 9 August 2024European AI Office ↗European Union · Official institutional resource · checked September 2026
KEEP THE CONVERSATION GOING
Media & workshop enquiries
A learning experience by Beatrix Meszaros, practical AI trainer and keynote speaker.
beatrix@pedey.co.uk ↗
Contact Beatrix through her website →
For interviews, facilitated sessions or questions about this educational project.