Breaking

Anthropic and OpenAI Warn Autonomous AI Self-Improvement Could Close Policy Window by 2028

September 28, 2026 at 11:09 EDT

Foundation Models
Policy & Regulation

AI pioneers at Anthropic and OpenAI are sounding alarms that the ability for humans to govern artificial intelligence development may narrow dramatically within the next two years, as systems increasingly automate their own improvement with minimal human oversight.

September 2026 · AI News

Race to Autonomous AI Self-Improvement

Frontier AI labs warn the window for human governance of AI development could narrow by 2028 as systems increasingly automate their own advancement.

80%+

Claude's production code

60%

Chance of autonomous RSI by 2028

8×

Code output per engineer

AI Research Timeline

Today

By 2027

By 2028

Research output accelerating: AI capability gains doubling every four months. Anthropic estimates 60% probability of fully autonomous self-improvement within two years.

→

Current: Humans direct research strategy

AI automates execution of specified tasks

→

Risk scenario: Full autonomy

AI formulates goals and iterates without human direction

Continue reading

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