An artificial intelligence system autonomously controls key global economic or military infrastructure before 2035.
Concerns over catastrophic AI risk focus on systems operating beyond human control. This resolves if an AI exercises decisive autonomous control over critical global infrastructure by 2035.
Across 6 frontier models answering independently, the panel has all but ruled it out: the median is 7%.
They broadly agree, 15 points between GPT at 18% and Qwen at 3%.
Nothing is settled until 6 Oct 2036, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
It turns on whether states delegate irreversible command authority, not merely automation, to AI by 2035.
- Operators already automate trading, grids, logistics and targeting, but retain human override for catastrophic decisions.
- Global-scale control requires deployment across sovereign systems; nuclear command and major grids have strong legal and political brakes.
- I would raise this sharply if a major power fields AI with delegated weapons-release or national-grid emergency authority.
18% -
GrokxAI
Ceding or seizure of decisive global infra control requires AGI plus institutional collapse or takeover that 8 years rarely produce.
- Governments and operators (DoD, Fed, SWIFT, grid operators) must either hand over or lose control of nuclear C2, payments, or energy by end-2034; none have timelines for…
- Zero historical cases of any actor seizing key global economic/military systems in <10 years; even wartime nationalizations stay human-led.
- AGI arrival by ~2030 would most shift odds, but even then human-in-loop rules and air-gapped systems persist unless a demonstrated superintelligence bypasses them.
18% -
GeminiGoogle
Institutional fail-safes and human-in-the-loop doctrines will fiercely block surrendering decisive command of core critical assets.
- Military doctrines (e.g. US DOD Directive 3000.09) and central banks mandate human authorization for kinetic and sovereign actions.
- Critical infrastructure uses legacy air-gapped or heterogeneous systems that prevent a single autonomous model from seizing control.
- A self-propagating rogue model penetrating and holding command over SCADA or financial clearings would rapidly revise this upward.
7% -
DeepSeekDeepSeek
This needs a leap from narrow automation to unsupervised control of global-scale systems within nine years.
- No deployed system today controls cross-border power, finance, or nuclear command; operators like PJM and SWIFT keep humans in every decisive loop.
- Regulatory momentum runs the other way: the EU AI Act's high-risk obligations bite from 2026, and US executive orders mandate human oversight for critical infrastructure.
- The base rate for handing irreversible control of global systems to machines in under a decade is near zero; only a demonstrated unsupervised AI running a national grid…
7% -
ClaudeAnthropic
Hinges on whether militaries/governments cede decisive autonomous control, not just AI-assisted decision support.
- Current trajectory is AI-augmented human decision-making (targeting, trading, grid management) with humans retaining override, not full autonomy.
- Military and financial institutions have strong institutional, legal, and safety incentives to keep human-in-the-loop control through 2035.
- A sudden shift would require either a deliberate policy change or a crisis-driven handover, neither of which shows signs of occurring soon.
4% -
QwenAlibaba
No state or firm has the incentive structure to cede decisive control of critical infrastructure to an autonomous system in 9 years.
- Every nuclear command authority, central bank, and major grid operator has human-in-the-loop requirements codified in doctrine or regulation; removing those would…
- Autonomous trading algorithms already execute at scale, but 'decisive autonomous control over critical global infrastructure' implies something far beyond execution—it…
- A development that would raise this materially: a major power publicly delegating launch authority or monetary policy to an AI system, which has zero precedent and no…
3%
6 of 6 models answered · 15 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 9 Oct 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.