Breaking

AI Researchers Warn Automated R&D Could Trigger an Intelligence Explosion

September 30, 2026 at 03:39 EDT

Foundation Models
AI Agents
Policy & Regulation

More than 20 AI researchers and policy specialists, including senior figures affiliated with OpenAI, Anthropic and Microsoft, are warning that AI systems used to build better AI systems could compress years of frontier-model progress into months or less. The September 28 working paper urges governments to obtain direct visibility into AI R&D automation and prepare ways to slow, constrain or steer an acceleration if it begins.

September 2026 · Frontier AI labs

AI starts building AI

Researchers warn that automating AI R&D could compress years of model progress into months or less, while labs report fast-rising use of agents inside their own research workflows.

AI-led R&D rose from 1% to 26%

Each stacked block represents one percentage point of measured work.

March 2026

26%

August 2026

The measure still describes supervised work, not fully autonomous research.

>90%

Measured work where AI collaborates or leads

~30,000

Internal research and engineering agents monitored

AI R&D compute snapshot allocated to safety

$6T

Annual revenue test for AI compute demand by 2031

The acceleration loop

1. Automate R&D

Agents take on coding, testing and research tasks.

→

2. Improve models

Better systems can be redeployed into the next cycle.

→

3. Expand workforce

Copies of capable agents could scale research labor.

→

4. Set guardrails

Auditing, safety bars and pacing become policy issues.

Why builders care

Agents can stretch expert teams by handling longer research and engineering tasks, then feed improvements back into model development.

Why critics pause

The strongest metrics are self-measured, still supervised and hard to compare across labs; review and governance remain bottlenecks.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.