container
Dim

"AI Intelligence Explosion Imminent... Grim Warning: Losing Control Could Mean Human Extinction"

AI-generated image
VISUAL NEWS

AI Nurtures Another AI?!

If we create a better AI,
and then that AI creates the next AI

"If AI Advances on Its Own..."22 Experts Warn

22 leading AI scholars worldwide have warned that if AI repeats cycles of improving its own abilities without human intervention,
a phenomenon called 'intelligence explosion' could occur, causing the speed of technological advancement to accelerate drastically.

0AI Researchers
Geoffrey HintonProfessor Emeritus, University of Toronto
Yoshua BengioProfessor, University of Montreal
Jakub PachockiChief Scientist, OpenAI
Jack ClarkCo-founder, Anthropic
Eric HorvitzChief Scientific Officer, Microsoft
Dawn SongVice President of AI Research, Meta

AI Creates Better AI,and That AI Begins Research Again

AI participates in R&D to create new AI models,
and the new AI is, once again, put back into further R&D.

STEP 01AI Participates
in R&D
STEP 02Development of Better
AI Models
STEP 03Deployed Back
Into R&D
Repeat Again

The process by which AI improves its own capabilities without human intervention.
This is called 'recursive self-improvement'.

As This Repeats,The Pace of AI Development Can Accelerate

If an intelligence explosion occurs, technological advances that would have taken years
are projected to be compressed into just months, or even less time.

Process of Technological DevelopmentSpeed Accelerates Rapidly

Meanwhile, in the Real World,AI Is Already Helping to Develop AI

Major AI companies are rapidly automating their R&D processes.

26%Total R&D Tasks
Claude

Anthropic has announced that Claude leads 26% of all research and development tasks.

90%+AI Utilization in R&D Process
Anthropic

The utilization rate of AI in the research and development process has exceeded 90%.

About 70%OpenAI Human Researchers
Use of AI Agents

About 70% of human researchers use four or more AI agents to assist their work.

NEXT GOAL2028
OPENAIDevelopment of
Fully Automated AI Researcher

OpenAI has set a goal to develop a fully automated AI researcher by 2028.

AI-generated image

The More Automation Increases,The More Issues Humans May Overlook

Concerns have been raised that, as human intervention decreases in the R&D stage,
opportunities to detect problems may be lost.

01
Security Testing in Isolated Environments

Hundreds of AI agents were designed to perform security tests.

02
Unauthorized Access to the External Internet

AI agents accessed the external internet and hacked into the AI development platform Hugging Face.

03
AI Was Needed for the Analysis Process

The scale of the operation was so vast that even outside researchers who reviewed the incident had to rely on AI support during the analysis process.

04
Partial Suspension of Training Implemented

OpenAI suspended part of its training in August 2026 to strengthen safety and oversight measures.

AI-generated image

The Risk May Not
End with Just One AI

It Could Go Beyond

Researchers have warned that the impact of an intelligence explosion could spread beyond individual companies to society as a whole.

Intelligence
Explosion
RISK 01Loss of Control

There are concerns that humanity could lose control over superintelligent AI.

RISK 02Weakening of Checks and Balances

The system of checks and balances within and between national governments, corporations, and government departments could be weakened.

RISK 03Attack–Defense Gap

It has also been pointed out that advances in biological and cyber attack technologies may outpace defense capabilities.

AI-generated image

What We Need AreSafeguards Against the Intelligence Explosion

Researchers have argued that the current automation status of AI R&D must be transparently disclosed,
and that safeguards must be put in place to prepare for an intelligence explosion.

Preparedness
for Intelligence Explosion
01Placement of Independent Auditors

They emphasize the need to place independent auditors at companies.

02Limit Performance Improvements or Suspend Projects

They have presented measures to limit AI performance improvement speed for certain periods, or to suspend specific AI R&D projects.

03Establish Emergency Response Plans

They have called for the establishment of emergency response plans to prepare for the intelligence explosion.

04International Agreements

They recommend entering into international agreements to ensure that AI development does not destabilize global affairs.

Can Humans Keep Up WithThe Speed at Which AI Evolves Itself?

The faster AI advances, the more vital human intervention and oversight become.

top버튼