“RSI could result in humans losing practical control over AI development.”Building standards for the next phase of AI
OpenAI, published September 21, 2026
Stop AI Recursive
Self Improvement.
If we let AI lead or automate the training and growth of its self we will lose human control.
0%of AI is already trained by AI.
Five months ago, it was 1%.
Five months ago, it was 1%.
Source shown in graphic: Anthropic R&D Automation Index v2026.07
OpenAI, published September 21, 2026
“Recursive self-improvement should be illegal. We shouldn’t even engage in this.”CNN interview
Professor Yoshua Bengio, AI pioneer
What RSI means
Recursive self-improvement (RSI) means AI helping to create more capable AI, which can then help create something more capable again.
This process may continue forever.
AI companies are perfectly able to operate, innovate quickly and make good products, services and profits without RSI. Just stop RSI.
Just stop RSI.
AI can support humanity in many areas when deployed safely. But we say it must never lead, automate, or independently advance its self. This is a red line for humanity.
Please Watch
Tell your government.
Choose your country, find your elected official and send them a clear message to stop RSI. It’s their job.
2,614actions taken
Select your country to see the official route for contacting your representative.
Select a country to continue