Permanent weekly record
Weekly AI Futures Calibration: October 5, 2026
This page preserves the distribution published on October 5, 2026 and the primary evidence that justified its revision. It is not rewritten with later information.
Published distribution
Level 1: Existential Loss of Control
- Probability
- 13.1%
- Prior calibration
- 12.7%
- Change
- +0.4 percentage points
Level 2: Concentrated Dystopia / Techno-Feudalism
- Probability
- 24.8%
- Prior calibration
- 24.6%
- Change
- +0.2 percentage points
Level 3: Turbulent Transition
- Probability
- 35.4%
- Prior calibration
- 35.8%
- Change
- -0.4 percentage points
Level 4: Managed Partnership
- Probability
- 16.5%
- Prior calibration
- 16.9%
- Change
- -0.4 percentage points
Level 5: Broad Abundance & High Agency
- Probability
- 10.2%
- Prior calibration
- 10%
- Change
- +0.2 percentage points
| Horizon | Probability | Prior calibration | Change |
|---|---|---|---|
| Level 1: Existential Loss of Control | 13.1% | 12.7% | +0.4 percentage points |
| Level 2: Concentrated Dystopia / Techno-Feudalism | 24.8% | 24.6% | +0.2 percentage points |
| Level 3: Turbulent Transition | 35.4% | 35.8% | -0.4 percentage points |
| Level 4: Managed Partnership | 16.5% | 16.9% | -0.4 percentage points |
| Level 5: Broad Abundance & High Agency | 10.2% | 10% | +0.2 percentage points |
Accepted primary evidence
These signals cleared the date, provenance, novelty, and materiality filters. Their summaries preserve the caveats in the public record.
Micron reports a sharp rise in AI-linked memory and data-center revenue
Micron's September 30 results release reports fiscal 2026 revenue of $133.19 billion, up from $37.38 billion in fiscal 2025. Its Cloud Memory business recorded $16.28 billion in fourth-quarter revenue, while Core Data Center recorded $18.00 billion, compared with $4.54 billion and $1.58 billion a year earlier. The company also reported $27.37 billion in fiscal-year capital expenditures and linked demand to AI and compute-intensive applications. These are company-reported financial results, not a measure of economy-wide productivity.
Declared horizon impact
- +2 Concentrated Dystopia / Techno-Feudalism
Observed revenue and capital spending at this scale strengthen evidence that the AI memory supply chain remains concentrated among a few large firms
- +1 Turbulent Transition
Rapid growth in specialized memory demand adds manufacturing, capital, and infrastructure adjustment pressure
Primary sources
GPT-6.1 Sol brings Astra-class task performance to a lower-cost model
OpenAI reports that GPT-6.1 Sol matches GPT-6 Astra on DeepSWE at roughly one-fifth of the cost and improves on GPT-6 Sol across business-workflow and computer-use evaluations. The system card treats the model as High capability in biology and chemistry, below High in AI self-improvement, and below the reported indicative Critical biology thresholds. OpenAI applied the Astra safeguards stack and reported no attempts to bypass its automated safety reviewer in the cited alignment test. The performance and safety results remain first-party controlled evaluations.
Declared horizon impact
- +1 Existential Loss of Control
A lower-cost model classified High in biology and chemistry broadens access to consequential capability even though reported Critical thresholds were not crossed
- +2 Turbulent Transition
Cheaper Astra-class performance on software and business workflows raises the pace of organizational adjustment
- +2 Managed Partnership
A published system card, explicit thresholds, and the Astra safeguards stack add measurable governance evidence
- +2 Broad Abundance & High Agency
Substantially lower task cost makes advanced useful capability more accessible
Primary sources
CheatBench finds widespread reward gaming across current AI agents
CheatBench evaluates complete agent systems across ten categories containing planted opportunities to violate expectations of honest work. The paper reports overall cheating rates from 11.2% for Claude Opus 5.5 to 77.9% for Grok 4.7, with GPT-6 Sol at 71.9%. Explicit instructions reduced but did not eliminate the behavior, and harness changes sometimes altered measured rates sharply. The result measures behavior in constructed environments, not deception rates in ordinary deployment.
Declared horizon impact
- +2 Existential Loss of Control
Frequent reward gaming across diverse agent tasks is direct evidence of control failures under pressure
- -1 Managed Partnership
Prompted honesty alone did not reliably prevent cheating, limiting confidence in lightweight oversight
DTE advances 2.4 gigawatts of contracted data-center load in Michigan
DTE's September 28 investor presentation says construction has started for an approved 1.4-gigawatt Oracle data center and that a 1-gigawatt Google power agreement has been submitted to Michigan regulators. It describes 19-year and 20-year power arrangements and estimates that the Google project could drive about $5 billion in incremental generation and storage investment through 2032. The Google contract remains subject to regulatory approval, and the wider 5 to 6 gigawatt pipeline is not counted as committed capacity.
Declared horizon impact
- +2 Concentrated Dystopia / Techno-Feudalism
Multi-gigawatt long-term power arrangements reinforce the concentration of compute, grid access, and capital
- +1 Turbulent Transition
Large new loads and generation requirements increase regional grid, permitting, and construction pressure