AI Risk Defusal & US-China Bilateral Treaty Sandbox

Bilateral Treaty Controls

Manipulate international constraints & inspect systemic impact

Scenario Presets

Compute Ceiling Cap 1.0e+26 FLOPs

Dual-use training hardware limits. Lower caps curb uncontrollable recursive breakout runs.

Shared Verification Inspectorate 75%

Joint US-China technical audit teams inspecting datacenters for clandestine mega-clusters.

Autonomous Weapon Ban Legally binding moratorium on AI target authorization without human-in-the-loop.

Open-Source Model Guardrails 85%

Cryptographic weight watermarking and CBRN-capability screening on released weights.

Nobel Peace Prize Incentive Vector 88.5 Index

Historic peacemaker incentive: Substantial mutual de-escalation presents unprecedented historic diplomatic legacy.

“Donald Trump is about to have the opportunity of a lifetime—if he is up for it... if Mr Trump were to help defuse AI’s most explosive risks, then the Nobel prize would be there for the taking.” — Gary Marcus, Guest Essay in The Economist

Global AI Threat & Stability Matrix

Interactive D3 force simulation: node tensions respond dynamically to parameter shifts

United States

China

Catastrophic Vector

Treaty Verification Hub

Nobel De-escalation Node

Catastrophic Risk 14.2% Decaying (-73% baseline)

Diplomatic Viability 88.5% Bilateral ratification odds

Inspection Coverage 75% Semiconductor verification

Geopolitical Posture Stabilized Crisis communication active

Understanding the hypothetical treaty score

Your browser does not support video.

Read the explanation

The treaty sandbox assigns two scores from checkbox and slider inputs. With the other defaults held fixed, inspector coverage contributes point three eight per point to viability and subtracts point two eight per point from the risk score. At seventy five, the assigned results are eighty eight point five viability and fourteen point two risk. The curves show the executable formulas, with clipping. These numbers are not measured diplomatic outcomes, validated catastrophe probabilities, or evidence that inspections or a treaty actually exist. The escalation preset sets inspector input to five, guardrails to fifteen, disables the ban, and raises the compute exponent to twenty eight. Its assigned risk is sixty two point one, not over eighty as the earlier explanation claimed. Viability is forty six point nine after the high-compute penalty. With the allowed input ranges, all risk terms subtract from sixty five, so this formula cannot exceed that baseline. The ninety four upper clamp never activates for these native controls. This is a toy scoring comparison, not a calibrated arms-race forecast. The status labeled ratified framework requires assigned viability at least eighty and risk at most twenty five. Otherwise viability at least sixty returns provisional accord; if that fails and risk is above fifty, it returns bilateral breakdown. The default and escalation points fall in different regions of this decision plane. A status string does not ratify an actual framework, create international agreement, or verify legal compliance. The region merely explains how this browser illustration names its states. The prize index combines sixty percent viability and forty percent risk complement. Defaults produce eighty seven point four two, displayed as eighty seven point four. It names values at least seventy eight high potential and at least fifty two moderate potential. These are author-assigned bins, not evidence of prize probability or a committee evaluation. Compute exponent values between twenty five and twenty seven leave viability unchanged, though they still change the risk term. Export saves a hypothetical brief. It does not submit diplomacy, verify current events, or establish any real Nobel candidacy.