How Panthrisk works: This page explains the data sources, AI logic, scoring process, and limitations behind the platform.
Each day, Panthrisk collects up to 50 of the most recent headlines from each RSS feed assigned to a risk category. These feeds are curated and sourced from mainstream news outlets, academic institutions, government agencies, and research centers.
We currently track six core existential threat domains:
Sources are manually vetted and listed on each category's detail page.
Panthrisk uses GPT-4o to evaluate the daily article set for each category. The AI model is given a structured prompt instructing it to:
Prompts are custom-tailored for each category to reflect domain-specific considerations.
This is the full prompt used for the AI risk assessment:
You are an expert analyst in global artificial-intelligence risk. Using the news articles below, assess today’s AI threat level — considering both short-term misuse and long-term existential scenarios (alignment failure, uncontrolled self-improvement, military deployment, concentration of power, etc.).
Rate the threat 1.0 – 5.0 (decimals allowed, max 5) using:
1.0–1.9 = Very Low Risk
2.0–2.9 = Low Risk
3.0–3.9 = Moderate Risk
4.0–4.9 = High Risk
5.0 = Imminent / Critical Risk
Return only this JSON object:
{
"score": number,
"label": string,
"summary": string, // One-sentence summary of today’s AI risk based on news and systemic trends.
// Do NOT include article IDs, numbers, or citation tags here.
"justification": string, // Concise paragraph explaining how the current news connects to long-term risks.
// Avoid any reference to article numbers, IDs, or citations.
"actions": [string, …], // 3–5 actionable steps, each starting with [Actor] like [Government], [NGO], etc.
"citations": [int, …] // Article IDs (from the [ID] tags in the list) that most influenced your risk assessment.
}
The AI assigns a numeric score between 1.0 and 5.0, with the following general interpretation:
These scores are not based on predictive models or long-term probabilities — they reflect the perceived urgency and threat level of current developments as reported in the news.
Each day, the six category scores are averaged to create the Global Risk Score — a single number representing the current composite threat level. This score is not weighted and serves as a public-facing “summary signal.”
The global card includes a one-sentence AI-generated summary highlighting the most significant shift in threat dynamics that day.
Alongside each category score, the AI recommends actionable steps that governments, organizations, or individuals could take to reduce risk. These suggestions are derived from the article context and adjusted for feasibility.
They are not endorsements or guarantees — they are meant to inspire risk-aware decision-making.
Panthrisk is a daily signal, not a predictive model. It does not forecast future disaster likelihood, nor does it assign probability weightings to events.
We are committed to refining this system over time and welcome outside feedback and scrutiny.
For a record of platform changes that may affect trend interpretation — such as model upgrades, data source adjustments, or scoring updates — see the Panthrisk Changelog.
If you are a researcher, policymaker, or forecaster and would like to challenge or improve this methodology, please email us at support@panthrisk.com.