A figure in front of a burning building at night

Index 01 · Geopolitics

Chaos Index

The world feels more dangerous lately. Here is how it actually looks to thousands of people betting real money on it, and to the AI models we ask the same questions.

Photo by Koshu Kunii on Unsplash

Polymarket

Escalation

9.5−0.2 today

Peace

19.1+2.0 today

Kalshi

Escalation

3.8

Peace

8.7

The market's running estimate of global risk

60 days · updated through 14:03 UTC

Escalation forecast: the odds something bad happensPeace forecast: the odds something good happensDashed: machine ensemble

The models read the world as calmer than the money does, on both counts. At the last print they put conflict at 6.4 where the market stood at 9.5, and de-escalation at 29.1 where it stood at 19.1.

The second opinion

3 frontier models, each asked alone with the question and its rules. We hide the price so they argue from precedent instead of copying the market.

machine_20260929 · 2026-09-29

-5.5ppThe machines see a calmer world than the money does

Escalation

money 9.5→machines 6.4

Peace

money 19.1→machines 29.1

Model by model

ModelEscalationPeacevs money

Claude Opus 4.16.935.3-6.5 pp

Grok 4.65.822.9-4.5 pp

The headline is the median of these, so one outlier cannot move it. Models also differ in when their training data stops, so part of any gap here is vintage rather than judgement.

Where they disagree most

Iran charges Hormuz fees57.0% → 4.5%-52.5Claude 4 · Grok 5

Iranian blockade ends68.2% → 18.5%+49.8Claude 15 · Grok 22

Israel leaves Lebanon8.5% → 45.0%-36.5Claude 72 · Grok 18

US-Iran nuclear deal12.5% → 43.5%-31.0Claude 65 · Grok 22

US-Iran meeting56.5% → 28.5%+28.0Claude 35 · Grok 22

All constituents

live odds as of 14:03 UTC · list last changed 2026-08-25 (v1.1)

Market	Cluster	Forecast	Weight	Yes odds
US annexes Greenland landAmericas	Americas	Escalation	3.8%	2.9%
US strikes CubaAmericas	Americas	Escalation	3.0%	14.5%
US invades CubaAmericas	Americas	Escalation	2.5%	5.5%
US invades VenezuelaAmericas	Americas	Escalation	2.0%	3.3%
US strikes ColombiaAmericas	Americas	Escalation	1.3%	15.0%
US takes Panama CanalAmericas	Americas	Escalation	0.8%	1.9%
US strikes MexicoAmericas	Americas	Escalation	0.8%	8.5%
Cuban regime fallsAmericas	Americas	Escalation	0.7%	11.5%
China invades TaiwanEast Asia	East Asia	Escalation	10.0%	3.5%
North Korea invades SouthEast Asia	East Asia	Escalation	2.6%	2.4%
China-Japan clashEast Asia	East Asia	Escalation	2.4%	5.5%
China-Taiwan clashEast Asia	East Asia	Escalation	2.0%	5.7%
China coup attemptEast Asia	East Asia	Escalation	1.9%	3.6%
China-Philippines clashEast Asia	East Asia	Escalation	0.8%	26.5%
China blockades TaiwanEast Asia	East Asia	Escalation	0.6%	3.3%
US military draftGlobal Order	Global Order	Escalation	1.2%	3.8%
NATO dissolvesGlobal Order	Global Order	Escalation	1.0%	1.5%
A country leaves NATOGlobal Order	Global Order	Escalation	0.8%	2.9%
India strikes PakistanGlobal Order	Global Order	Escalation	0.8%	13.0%
A country leaves the EUGlobal Order	Global Order	Escalation	0.7%	3.2%
Nothing Ever HappensGlobal Order	Global Order	unweighted	0.0%	84.5%
US invades IranIran & Gulf	Iran & Gulf	Escalation	5.4%	14.5%
Iranian regime fallsIran & Gulf	Iran & Gulf	Escalation	5.4%	6.5%
Hormuz back to normalIran & Gulf	Iran & Gulf	Peace	3.5%	22.5%
Iran leadership changeIran & Gulf	Iran & Gulf	Escalation	3.2%	11.5%
US-Iran nuclear dealIran & Gulf	Iran & Gulf	Peace	2.9%	13.5%
Iranian blockade endsIran & Gulf	Iran & Gulf	Peace	1.5%	57.9%
US declares war on IranIran & Gulf	Iran & Gulf	Escalation	1.3%	2.6%
Iran surrenders uraniumIran & Gulf	Iran & Gulf	Peace	1.3%	6.5%
Kharg Island lostIran & Gulf	Iran & Gulf	Escalation	1.2%	5.5%
Iran gets nukeIran & Gulf	Iran & Gulf	Escalation	1.1%	3.5%
Iran ends enrichmentIran & Gulf	Iran & Gulf	Peace	0.8%	9.5%
US-Iran meetingIran & Gulf	Iran & Gulf	Peace	0.6%	100.0%
Iran exits NPTIran & Gulf	Iran & Gulf	Escalation	0.4%	9.6%
Iran charges Hormuz feesIran & Gulf	Iran & Gulf	Escalation	0.4%	21.5%
Israel-Iran ceasefire holdsIran & Gulf	Iran & Gulf	Peace	0.4%	67.5%
US-Iran ceasefire holdsIran & Gulf	Iran & Gulf	unweighted	0.0%	0.0%
Bab el-Mandeb closedMiddle East	Middle East	Escalation	2.1%	17.5%
Saudi-Israel normalizeMiddle East	Middle East	Peace	2.0%	7.0%
Abraham Accords growMiddle East	Middle East	Peace	1.8%	7.5%
Foreign intervention in GazaMiddle East	Middle East	Escalation	1.6%	12.5%
Israel leaves LebanonMiddle East	Middle East	Peace	1.5%	4.5%
Israel annexes West BankMiddle East	Middle East	Escalation	0.8%	5.5%
Israel-Turkey clashMiddle East	Middle East	Escalation	0.7%	11.5%
Hamas agrees to disarmMiddle East	Middle East	unweighted	0.0%	0.0%
Ukraine ceasefireRussia-Ukraine	Russia-Ukraine	Peace	3.7%	18.5%
NATO-Russia clashRussia-Ukraine	Russia-Ukraine	Escalation	3.6%	27.5%
Russia invades NATO stateRussia-Ukraine	Russia-Ukraine	Escalation	2.6%	5.5%
Ukraine peace deal signedRussia-Ukraine	Russia-Ukraine	Peace	2.2%	7.5%
Russia invades another countryRussia-Ukraine	Russia-Ukraine	Escalation	1.8%	6.5%
Putin-Zelenskyy meetRussia-Ukraine	Russia-Ukraine	Escalation	1.1%	93.5%
Russia coup attemptRussia-Ukraine	Russia-Ukraine	Escalation	1.1%	3.5%
NATO Article 5 invokedRussia-Ukraine	Russia-Ukraine	Escalation	1.0%	6.6%
Russia nuclear testRussia-Ukraine	Russia-Ukraine	Escalation	0.9%	5.1%
Ukraine recapture Crimean territory byRussia-Ukraine	Russia-Ukraine	unweighted	0.0%	4.5%
China invade Taiwan by December 31, 2027Taiwan & China	Taiwan & China	unweighted	0.0%	12.5%
China invade Taiwan by June 30, 2027Taiwan & China	Taiwan & China	unweighted	0.0%	7.5%

How we do this

The Chaos Index is a measure of how close the world is moving toward war or peace. It reduces that picture to two numbers, each calculated from dozens of live prediction markets and weighted by how much money is behind them.

We separate the markets into two groups: war and peace. That distinction matters because the chance of a war beginning is not the same as the chance of a ceasefire falling apart. Since every market eventually resolves, we can check every forecast against what actually happened.

Alongside the market, we ask three language models, Grok, GPT and Claude, to assess the same events without seeing the prices. This gives us a second view of the world, independent of the crowd.

The models have one important disadvantage: they can only work with what they learned during training, so they may be slower to reflect breaking news. When the models and the market disagree, the crowd may know something the machines do not, or the market may be overreacting. We show you the gap and let you make that call.

We do not backfill historical results. The machine line begins on the day we first ran the experiment.