AI Risk & AGI Progress Index

Every number on this page is a cumulative counter of news pressure, produced by an automated pipeline — no editorial opinion involved:

• Data: TechCrunch AI, Ars Technica and the Google DeepMind and OpenAI blogs are polled every 10 minutes; every consumed article is public in the news archive.
• Analysis: a large language model (currently Anthropic's Claude Opus 5.5; the model has changed several times since 2025) scores each article from −100 to +100 on four axes — control-loss probability and timing, AGI progress and timing — with a published justification on every article page.
• Index: each score becomes a small move through fixed divisors, and the moves add up with nothing decaying — so the numbers mostly climb, because most news is scored in the alarming direction.

Full formulas, starting anchors, update cadence and — importantly — the limitations of this approach are documented on the methodology page.