About ww3yet
What is ww3yet?
ww3yet is a live global conflict risk index powered by frontier artificial intelligence. Every evening at 18:00 CET, we query 12 advanced AI models to assess the current state of global geopolitical stability. Each model independently evaluates verified open-source intelligence from 9 curated news and research sources, then returns a risk score on a 0–100 scale. The result is not a prediction or speculation—it is a transparent, data-driven consensus of what the world's most capable AI systems believe about conflict risk today. The index updates daily, providing a real-time snapshot of global tensions grounded in factual reporting.
How It Works
Our daily pipeline distributes the same structured geopolitical prompt to 12 frontier AI models via OpenRouter. The models queried include GPT-5.4 from OpenAI, Claude Sonnet 4.6 from Anthropic, Gemini 3.1 Pro from Google, Grok 4 from xAI, GLM-5 from Zhipu AI (China), Mistral Large from Mistral (France), Qwen 3.5 397B from Alibaba (China), DeepSeek R1 from DeepSeek (China), MiniMax M2.5 from MiniMax (China), Command R+ from Cohere (Canada), Llama 4 Maverick from Meta (USA), and Solar Pro 3 from Upstage (South Korea). Each model receives the same context: curated articles from Reuters, Associated Press, BBC News, Al Jazeera, Defense One, The War Zone, Bulletin of the Atomic Scientists, SIPRI, and Arms Control Association. Models return a risk score (0–100), confidence level, key escalating and de-escalating factors, and detailed reasoning. This distributed approach ensures no single model's bias dominates the assessment.
AI Consensus Scoring
Individual model scores are aggregated using a weighted median approach, which ensures that no single model's outlier assessment distorts the final result. The median is more robust than a simple average because it is resistant to extreme values. Each model's score is weighted based on historical accuracy and calibration, so consistently reliable models have greater influence on the consensus. The final consensus score is persisted to our database and published on the site each evening at 18:00 CET. This methodology reflects the principle that AI consensus is more trustworthy than any single model's judgment, and more transparent than human editorial decision-making alone.
Our Data Sources
We monitor 9 carefully selected sources across four categories. Wire services—Reuters, Associated Press, and BBC News—provide real-time breaking news from conflict zones and diplomatic developments across 200+ locations worldwide. Defense publications—Defense One and The War Zone—deliver military analysis, OSINT verification, and weapons systems reporting. Research institutes—Bulletin of the Atomic Scientists, SIPRI, and Arms Control Association—provide expert analysis on nuclear risk, arms transfers, military spending, and disarmament negotiations. Regional outlets like Al Jazeera ensure perspectives from the Middle East and Global South are not underweighted. Each source is scanned every 6 hours for geopolitical signals, and articles matching our conflict keywords are fed to the AI models as context.
The Risk Scale
The ww3yet risk scale runs from 0 to 100, with five distinct risk levels. Scores of 0–20 indicate Low Risk, reflecting stable global conditions with no imminent threats. Scores of 21–40 indicate Guarded conditions, with elevated tensions in isolated regions but no systemic instability. Scores of 41–60 indicate Elevated Risk, with significant geopolitical instability and multiple concurrent crises. Scores of 61–80 indicate High Risk, with major international crises, military mobilizations, or alliance fractures. Scores of 81–100 indicate Severe Risk, reflecting conditions consistent with the onset of global conflict. The current score and full model breakdown are published publicly on the ww3yet homepage, updated each evening.
Why AI Consensus?
No single artificial intelligence model has perfect judgment. Each model has different training data, architectural biases, and blind spots. By querying 12 models from 12 different providers across 5 countries — OpenAI, Anthropic, Google, and xAI (USA), Zhipu AI, Alibaba, DeepSeek, and MiniMax (China), Mistral (France), Cohere (Canada), and Upstage (South Korea) — we mitigate individual biases and create a more robust assessment. A model trained primarily on Western sources might overweight NATO perspectives; a model trained on Chinese data might overweight Beijing's interests. Chinese models may downplay risks involving Taiwan or Xinjiang; Western models may overweight NATO perspectives. With 4 Chinese and 5 American models plus French, Canadian, and Korean perspectives, the consensus balances Eastern and Western viewpoints into a more neutral assessment. This is more transparent than relying on a single proprietary model or human analysts, and more accurate than any individual model alone.
Disclaimer
ww3yet is an informational tool designed to provide transparency into AI consensus on global conflict risk. It is not financial advice, military intelligence, or policy guidance. The risk score reflects the consensus of multiple AI models based on publicly available information from verified news and research sources. ww3yet is operated by Unique (Deutschland) GmbH, Hamburg, Germany. Users should consult official government sources, defense analysts, and domain experts before making decisions based on our risk index. The index is updated daily and reflects conditions as of the publication date.