Best crypto insights delivered straight to your inbox.
AI systems might be able to rebuild themselves from 2028, says Anthropic co-founder
- Anthropic’s Clark says AI systems might be capable of rebuilding themselves from 2028.
- He noted most models already have what it takes to fully automate end-to-end R&D.
- “I don’t know how to wrap my head around it,” Clark said.
Anthropic’s co-founder Jack Clark has made a bold prediction that AI systems may be capable of rebuilding themselves from 2028. “I’m not sure society is ready,” he said.
I’ve spent the past few weeks reading 100s of public data sources about AI development. I now believe that recursive self-improvement has a 60% chance of happening by the end of 2028. In other words, AI systems might soon be capable of building themselves.
— Jack Clark (@jackclarkSF) May 4, 2026
AI systems now need less human oversight
“This is a big deal,” says Anthropic’s Jack Clark
New post: on Jan 14, I predicted that SWE time horizon by EOY would be ~24 hours. Now I think it’ll be >100 hours, and maybe unbounded. For the first time, I don’t see solid evidence against AI R&D automation *this year.* Link below. pic.twitter.com/NcP1HlZana
— Ajeya Cotra (@ajeya_cotra) March 5, 2026
Still letting the bank keep the best part? Watch our free video on being your own bank.
FAQs
Who is Jack Clark and why does his view on AI matter?
Jack Clark is the co-founder of Anthropic, one of the leading frontier AI companies, and the author of Import AI, a newsletter covering AI research that has run for over 450 issues. His position inside a major lab and his track record of analyzing public research give his forecasts weight in the AI policy and research community.
What does "recursive self-improvement" mean in this context?
According to Clark's essay, it refers to an AI system powerful enough to autonomously conduct its own research and development and build its own successor without human involvement. He estimates there is a 60% or greater chance this capability emerges by the end of 2028.
What evidence does Clark cite for his prediction?
Clark pointed to benchmark results including SWE-Bench, where top AI models went from 2% to 93.9% success on real-world coding tasks between 2023 and 2026, and METR data showing AI systems can now work independently on tasks for approximately 12 hours, up from 30 seconds in 2022.
CRASH COURSE
- Which cryptocurrencies can make you money
- How to boost your security with a wallet (and which ones are actually worth using)
- Little-known investment strategies that the pros use
- How to get started investing in crypto (which exchanges to use, the best crypto to buy etc)