How it works
Escalation Watch is an automated early-warning read on the risks that could lead to a third world war or a catastrophe that threatens civilization. It is meant to be realistic, not alarmist: it should stay calm about rhetoric and noise, and move only when concrete, corroborated events justify it.
What the global level means
| Level | Rough meaning |
|---|---|
| 1-2 | Great powers at peace and cooperating |
| 3 | Normal great-power rivalry; regional wars that do not involve great powers directly (most of the 2010s) |
| 4 | A serious regional war or crisis with great powers backing opposite sides |
| 5 | A major war involving a nuclear power, with deterrence holding (the 2022-2024 Ukraine war) |
| 6 | Several such wars or crises at once, or one great power fighting directly |
| 7 | Limited direct clashes between nuclear powers, or nuclear threats backed by force movements (like the 1983 Able Archer scare) |
| 8 | Sustained direct combat between great powers, or a nuclear crisis like Cuba 1962 |
| 9 | Nuclear use, or it appears imminent |
| 10 | General nuclear war or an unfolding extinction-level event |
Pandemics and AI or biological catastrophes are placed on the same scale by their potential to end civilization. Economic and political stress are treated as context that can accelerate a crisis, not as threats to civilization by themselves.
How each update is built
- Collect. Every hour, the system reads major world news feeds, targeted news searches and official sources (listed below). When the level is above 6, it checks every 15 minutes.
- Group into events. Headlines about the same thing are grouped into one event, so a story repeated by forty outlets counts as one event, not forty.
- Grade the evidence. Each event is graded by how many independent outlets carry it and how reliable they are. Wire services, papers of record and official sources count most. Single-outlet stories, tabloids and state-run media count for little, and state media is flagged because it reflects its government's line. Headlines that report statements, threats or speculation count less than reports of actions.
- Analyze. A skeptical AI analyst reviews the events for each category, judges each one on its own and against the others, and explains what it weighed and what it set aside. Its read is blended 50/50 with the measured data score.
- Check the tripwires. Fourteen concrete warning signs (for example direct NATO-Russia combat, a Taiwan blockade, a change in nuclear alert status, or a novel pathogen spreading between people in several countries) are checked against confirmed events. A tripwire can only be marked triggered by an action confirmed by several reliable outlets; this is enforced in code, not left to the AI.
- Global read. The analyst then writes one connected assessment and level, including why it is not higher and why it is not lower. That level is kept within 2.5 points of a weighted data composite and averaged over the last three reads, so a single run cannot swing it wildly.
- Hard rule. If the separate nuclear detonation watch (USGS seismic data plus multi-outlet news) detects a detonation, the level is set to at least 9 immediately.
Sources
News: BBC, Al Jazeera, New York Times, Guardian, Deutsche Welle, Defense News, Military Times, CNBC, and Google News searches across many outlets. Specialist: IAEA, Arms Control Association, WHO Disease Outbreak News, CIDRAP, ECDC, CDC Health Alert Network, BleepingComputer, The Record, Krebs on Security, The Hacker News, and AI coverage from the Guardian, MIT Technology Review and TechCrunch. Official data: US State Department travel advisories, the Federal Register, DHS terrorism advisories, FAA airspace status, CISA and NVD vulnerability data, USGS earthquake data, FRED economic data and market quotes. The Doomsday Clock setting is reported as published by the Bulletin of the Atomic Scientists.
Headlines are shown as short titles that link to the original article. Read the source for the full story.
Limits
This is an automated summary of public reporting, built by one person as a personal project. It is not an official warning system and not intelligence analysis.
- It only knows what is reported in public sources, and reporting can be wrong, late or slanted.
- AI analysis can misread events, miss context or be inconsistent between runs. The method is designed to reduce this, not to eliminate it.
- Small hour-to-hour moves of a few tenths are normal variation, not news.
- In an emergency, follow your local authorities and official alerts.
Escalation Watch is not affiliated with the Bulletin of the Atomic Scientists, any news outlet or any government.