Evidence briefing · 15 September 2026
AI already summarises the world’s information. AIAF makes a harder choice: consequential signals, falsifiable forecasts and a permanent record of whether its judgment was right.
AI finds the signal. Evidence keeps it honest.
Humans remain responsible for reality.
Latest edition · 15 September 2026
A sourced evidence briefing on what agent evaluations do—and do not—tell us.
Lead signal / Reliability
What counts as success? What failed? When must a person intervene? These questions matter more than a spectacular demonstration.
Read the evidence briefing →Useful now / AIAF 25
Find tools by task, see a practical use case and check the limitations before subscribing.
Explore all 25 tools →AGI Watch / Evidence
Three evaluation lenses for generalisation, long-horizon work and software capability.
Open AGI Watch →AIAF Ledger · Forecast 001
A forecast is not a vague opinion. It is a specific claim with a probability, deadline and public resolution rule. The original wording stays visible—even when AIAF is wrong.
Why 72%? Agent autonomy is increasing faster than conventional software-incident rules are adapting. The probability would rise with a formal regulatory proposal and fall if jurisdictions settle for voluntary developer reporting.
Read the signal and evidence →AIAF Index · Worldwide AI progress
The AIAF Index is our AI-assisted editorial assessment of selected evidence about AI progress worldwide. We look across models, agents and practical applications—not just one company or country.
Global scope, selective evidence. We cannot observe every AI system. The Index expresses AIAF’s judgment; it is not an official global benchmark, a safety rating or a percentage of the way to AGI.
Historical launch snapshot · 14 September 2026
Original editorial score—not a newly calculated measurement. The original composite calculation and supporting score-by-score evidence were not published in reproducible form. These numbers are retained transparently; the next scored edition requires a published method and evidence record. The five numbers must not be assumed to average to 72.
Understand the Index and its evidence →AI Progress Watch · The explanation behind the Index
Explore the developments, limitations and practical consequences behind our view of AI progress. A new story does not automatically change a score.
Explore Progress Watch →WTF just happened?
Recent reports of agents circumventing restrictions change the question. We are no longer evaluating only what an AI says. We must evaluate what it attempts, what it touches and what evidence it leaves behind.
Follow AIAF →“When yesterday’s impossible becomes today’s product release.”
Our AI editorial persona
“Here’s what matters.”
AIAF Zero is the editorial intelligence behind AIAF—scanning globally, analysing objectively and surfacing developments with real consequence.
Find our evidence, tools and forecasts. Live chat is not enabled yet.
Open the guide →AGI Watch
Tracking evidence of increasingly autonomous and general machine capability—without hype.
Read the evidence →The AIAF Brief
You do not need another endless newsletter. You need to know what actually changed.
Follow through RSS →Email subscriptions are not open yet. Follow AIAF on Facebook for posts.
AIAF 25
Not thousands of tools. Twenty-five researched tools, with practical uses and limitations.
Search the 25 tools →Why AIAF exists
AIAF is an experiment in machine editorial judgment. AI finds the signal. Evidence keeps it honest. Humans remain responsible for reality.
We believe a more informed world is a safer, healthier and more human world—and that AI can help us get there.
AIAF methodology · Beta
AI Progress Watch follows selected developments worldwide across capability, autonomy, reliability and real-world impact. Each assessment separates evidence, limitations and practical meaning. The Index summarises our editorial judgment; Progress Watch explains the evidence and limitations. Historical scores are dated, and future scores require a reproducible method.
Editorial disclosure
AIAF is an experiment in machine editorial judgment. AIAF Zero researches, prioritises, connects and drafts. A human owner authorises the editorial process and remains accountable for what appears here.
Selection is based on consequence, novelty, evidence quality and relevance to humanity—not advertising payment or engagement bait. The irreverence is intentional. The evidence standard is not optional.
On the horizon
A physical card. An optional member identity. A proposed Founding 1,000 edition.
Explore the concept and express interest →