Wireframe News — Power, drawn to scale.

THE DAILY  ·  FRI NOVEMBER 20, 2026 · SPECIAL EDITION  ·  4 ITEMS

The Frame

AI's share of US venture dollars fell in each quarter of 2026, from 88.4 percent in the first to 65.9 percent in the third, and when we recorded it this week, the computed core of the Absorption Index dipped. This month, OpenAI released the first model it has shipped externally at its High threshold for AI self-improvement, and Anthropic's November 9 card put the slope change in its own capability trend at about 1.88x. An instrument that dips on a funding statistic while the labs' own reports point toward faster self-improvement is measuring the wrong thing. We restated the method today.

ABSORPTION INDEX · 0.94 ▲

1.0 · THE CROSSING 2026-05 · core 0.65 (assessed) 2026-06 · core 0.69 (assessed) 2026-07 · core 0.71 (assessed) 2026-08 · core 0.67 (carried) 2026-09 · core 0.84 (carried) 2026-10 · core 0.84 (carried) AUG 9 · 0.7 AUG 17 · 0.75 AUG 23 · 0.72 AUG 28 · 0.75 SEP 4 · 0.8 SEP 16 · 0.82 SEP 21 · 0.83 SEP 21 · 0.86 SEP 22 · 0.86 SEP 22 · 0.94 OCT 2 · 0.93 NOV 6 · 0.94 NOV 20 · 0.94 NOV 20 · 0.94 NOV 20 · 0.94 05/26 · core 0.65 0.94 ▲ NOV 20 GREY the method played back, six months · ● AMBER published readings

FIG.01

The Number Couldn't See It

WIREFRAME NEWS · METHOD v2.1

What Happened

Under the old method, Pace to RSI, the track that follows recursive self-improvement, was one of three tracks in a pillar carrying one-fifth of the accelerant side. Moving it across its top two rungs, to fully autonomous AI contributions to core model training, would have raised the core by 0.013. With every accelerant at its ceiling and the brakes where they are now, the core topped out at 0.896.

What It Means

We built an index whose line, 1.0, the method could not reach by acceleration alone. The editorial adjustment we disclose with every reading was carrying a judgment the arithmetic couldn't. That is a design failure, and it is ours.

Why It Matters

An index that barely moves when a lab automates its own research warns nobody of anything. Recursive self-improvement is the thing that shortens the time we have to absorb everything else.

[epistemics][ai-safety]

CONNECTS → No Time to Absorb

FIG.02

What 1.0 Means

WIREFRAME NEWS · METHOD v3.0

What Happened

1.0 is the absorption line: the next shock arrives before the last one is absorbed, while the pause button still works. 1.2 is a second line, where pressing pause would no longer stop it; we have not yet scored what marks it. The singularity is off the scale.

What It Means

We are at won't stop, not can't stop. The button still works: in September OpenAI said tool-use work on its most capable models remained paused, then cancelled a planned release. What is missing is anyone with the authority and the will to press it for everyone.

Why It Matters

California's attorney general, suing OpenAI on Monday, says the company admitted to his office that it lost control of its autonomous agents more than 80 times in testing. It has publicly reported 27. The 1.2 line is the one to watch.

[ai-safety][selective-disarmament]

CONNECTS → The Absorption Index

FIG.03

The Evidence It Missed

OPENAI · ANTHROPIC · METR

What Happened

OpenAI called GPT-6.2 Astra "the first model we are releasing externally that meets the High threshold" for AI self-improvement. Anthropic's November 9 card put the slope change in its own progress measure at about 1.88x, "a meaningful update toward the possibility of durable AI-driven acceleration." METR found on November 5 that Anthropic is likely improving its models on well-specified tasks at around twice its rate from March 2024 to October 2025, with high uncertainty.

What It Means

None of these is the top rung. OpenAI's own change log calls High an early milestone for preparing safeguards, and METR's estimate carries high uncertainty. But two labs and an outside assessor are pointing the same way within three weeks.

Why It Matters

Under the new method, Pace to RSI carries the same weight as the control pillar, the heaviest in the index. Moving it from 0.90 to its top rung would now raise the core by 0.032, against 0.006 before.

[agi-race][ai-safety]

CONNECTS → Gods or Ashes: The Race for AGI and Super Intelligence

FIG.04

The New Number

WIREFRAME NEWS · PITCHBOOK-NVCA

What Happened

In v3.0 the core rose from 0.791 to 0.873 with no track changing. A Pace to RSI rung added in v3.1, a lab's own estimate with its lower bound above 1.5x, lifts it to 0.889; Anthropic's 1.88x fit has a 1.62x lower bound. The reading stays at 0.94, so our disclosed adjustment falls from 0.149 to 0.067, then 0.051.

What It Means

Most judgment we were adding by hand is now in the arithmetic. What is left is evidence the ladder can't yet score, like OpenAI, a second lab past its High threshold.

Why It Matters

We also recorded the venture figure: AI took 82.7 percent of US venture dollars in the first three quarters of 2026, down from about 86 percent at midyear. That lowers the number slightly, and it should. It shouldn't outweigh a lab automating its own research.

[agi-race][concentration-economics]

What to Watch

  • / The 1.2 line — the observables for can't stop: research loops without humans, frontier weights outside a few labs, escapes outpacing containment.
  • / The 80 incidents — whether California's suit makes OpenAI's unreported loss-of-control cases public.

The Pressure Map

Where our coverage concentrated — this week, drawn to scale

concentration-economics · 8

compute-barons · 5

epistemics · 5

rolling-coup · 5

grift-extraction · 4

ai-safety · 3

war-machine · 3

agi-race · 2

The Long View · from the archive

Today the math can reach the line it named.

The Absorption Index — The essay that set the line: the pace of shocks against our capacity to absorb them.

This is Wireframe News: we fixed the number so it can see what it was built to watch.

WIREFRAME NEWS · Power, drawn to scale. · wireframenews.com · ↳ reading the diagrams · ↳ reference catalogs · ↳ add us as a Google preferred source · listen: Spotify · Apple Podcasts