We\u2019ve seen it in films and pop culture for decades; it looks like sci-fi is no longer fiction. Artificial intelligence systems are starting to help build the next generation of AI models, according to new research released by Anthropic. The company says the trend could eventually lead to AI systems designing and improving themselves with minimal human input.<\/p>\n
Anthropic outlined the warning in a new blog post from its research-focused Anthropic Institute. The company said the industry may move toward \u201crecursive self-improvement\u201d sooner than many governments and institutions expect.<\/p>\n
The concept describes a future where one AI model develops the next version of itself. Researchers still guide the process today. However, Anthropic said AI already handles a growing share of coding, debugging, and technical research inside the company.<\/p>\n
Faster AI development<\/p>\n
Anthropic pointed to internal data showing how rapidly AI tools now contribute to software engineering work. The company said Claude-generated code accounts for more than 80% of the code merged into Anthropic\u2019s systems as of May 2026. Before the launch of Claude Code in early 2025, that figure sat in the low single digits.<\/p>\n
Our internal data shows Claude is accelerating AI development\u2014a possible path to recursive self-improvement, or AI autonomously building a more capable successor.<\/p>\n
It\u2019s happening faster than we thought, and the implications deserve greater attention. https:\/\/t.co\/OVVPJO7VQx<\/a><\/p>\n \u2014 Anthropic (@AnthropicAI) June 4, 2026<\/a><\/p>\n The company also said engineering productivity has surged alongside those changes. Anthropic engineers now merge roughly eight times more code per day than they did in 2024.<\/p>\n Jack Clark, Anthropic\u2019s co-founder and head of policy, said the company wants lawmakers and institutions to understand what may come next. \u201cWe\u2019ve always found that the best thing to do is to socialize the concept and basically give people a sense of what\u2019s coming,\u201d Clark said in a release<\/a>. <\/p>\n Clark added that AI progress appears to be accelerating instead of slowing down. He said the shift could drive major gains in medicine, science, and other technical fields.<\/p>\n Benchmarks moving rapidly<\/p>\n Anthropic also highlighted public benchmarks that track AI performance across software engineering and scientific research tasks.<\/p>\n The company said AI systems now complete increasingly complex assignments over longer periods without human intervention. Anthropic claimed the length of tasks models can reliably handle has doubled roughly every four months.<\/p>\n According to the company, Claude Opus 3 completed coding tasks lasting only minutes in early 2024. A year later, Claude Sonnet 3.7 managed work that required about 90 minutes. Anthropic said Claude<\/a> Opus 4.6 later handled assignments lasting up to 12 hours.<\/p>\n The company also referenced SWE-bench, a software engineering benchmark that tests whether AI can fix real-world coding issues inside open-source projects. Anthropic said frontier models moved from weak scores to near-saturation on the benchmark within two years.<\/p>\n Another benchmark, CORE-Bench, measures whether AI can reproduce published scientific research. Anthropic said AI systems improved from reproducing results roughly 20% of the time in 2024 to near-perfect performance about 15 months later.<\/p>\n Risks and oversight<\/p>\n Anthropic stressed that major gaps still separate current systems from fully autonomous AI development. The company said humans continue to define goals, judge results, and decide which research directions matter most.<\/p>\n Still, the company warned that stronger autonomous systems could create new risks if oversight tools fail to keep pace.<\/p>\n \u201cAs organizations, and eventually probably as societies, we need to figure out the tools to validate and verify\u201d AI-generated work, Clark said. He added that future systems must remain aligned with human goals and public interests.<\/p>\n