The AI Overlord Index · October 2026 issue
Projected end of human control: November 2031
Our monthly index of how far AI has taken over reads 39.4 out of 100, up 1.6 points in September. That brings the projected date about 7 months closer than last month's estimate of June 2032. Built from 12 public measures, from sources including GitHub, Epoch AI, the US Census Bureau, federal court dockets and SEC filings.
The index rose to 39.4 from 37.8, led by what AI can do and what AI does on its own, while the brakes held at 38.
Twelve measures from thirteen outside data sources, run by eleven organisations, in five parts, plus a brakes gauge. 0 is where each stood on November 30, 2022, the day ChatGPT launched; 100 is the point where AI would effectively be in charge of it. Parts are combined with a simple weighted average.
The AI Overlord Index every month since day zero, after brakes
Measures join when their data begins (OpenRouter January 2025, the current Census question November 2025). With weights moved up to 10 points, 80% of the results fall between 36.7 and 42.4.
Show the numbers: the index, last 13 months
| Month | Index | Before brakes | Brakes gauge |
|---|---|---|---|
| Sep 2026 | 39.4 | 44.5 | 38.0 |
| Aug 2026 | 37.8 | 42.7 | 38.0 |
| Jul 2026 | 37.4 | 42.3 | 38.0 |
| Jun 2026 | 35.9 | 40.7 | 39.0 |
| May 2026 | 35.1 | 39.8 | 39.0 |
| Apr 2026 | 33.9 | 38.4 | 39.0 |
| Mar 2026 | 32.7 | 37.1 | 39.0 |
| Feb 2026 | 31.0 | 35.1 | 39.0 |
| Jan 2026 | 30.1 | 34.1 | 39.0 |
| Dec 2025 | 30.2 | 33.3 | 31.7 |
| Nov 2025 | 29.3 | 32.4 | 31.7 |
| Oct 2025 | 29.0 | 32.0 | 31.7 |
| Sep 2025 | 27.7 | 30.4 | 28.9 |
This month
What moved in September
| Part, and its lead finding | Weight | August | September | Change |
|---|---|---|---|---|
| What AI does on its ownAI agents opened 1.4 million code changes on GitHub in September | 32% | 60.5 | 62.9 | ▲ 2.4 |
| What AI can doThe best AI model scores 169.2 on Epoch's capability index | 22% | 47.4 | 51.0 | ▲ 3.6 |
| How much we lean on itMore than 1 in 5 US businesses now use AI | 20% | 24.7 | 24.9 | ▲ 0.2 |
| What AI brokeCourts have caught AI-made-up material 2,211 times since ChatGPT launched | 14% | 13.1 | 14.5 | ▲ 1.4 |
| Money and computeHeld: no change from August | 12% | 50.9 | 50.9 | no change |
| The index before brakes | 100% | 42.7 | 44.5 | ▲ 1.8 |
| Brakes gaugeBinding AI rules in force; can take off up to 30% | 38.0 | 38.0 | no change | |
| The AI Overlord Index | 37.8 | 39.4 | ▲ 1.6 |
How the 39.4 is made: each part's score times its weight (What AI does on its own 62.9 times 32% = 20.1; What AI can do 51.0 times 22% = 11.2; How much we lean on it 24.9 times 20% = 5.0; What AI broke 14.5 times 14% = 2.0; Money and compute 50.9 times 12% = 6.1). Added together, that is 44.5 before brakes (the five numbers are rounded, so as shown they add to 44.4). The brakes gauge is at 38.0 out of 100 and can take off up to 30%, so this month it takes off 11.4% of 44.5, which is 5.1. 44.5 minus 5.1 = 39.4.
- Which of the 12 measures went up or down, and why
- The AI incidents that mattered in September
- The AI rules now in force, and how much they slow things down
- How the number is made, so you can check it