- Anthropic says Claude now writes much of the code used to develop its own AI systems, boosting productivity.
- The company says AI capabilities are growing faster, with task complexity doubling about every four months.
- Anthropic warns AI could eventually help create better AI, making oversight and safety more challenging.
Artificial intelligence company Anthropic says AI is improving so quickly that it is starting to help create the next generation of AI systems. The company warns that this could eventually make it harder for humans to maintain control over AI development.
In a new research paper, Anthropic discussed the possibility of “recursive self-improvement.” This is a situation where AI systems design, build, and train more advanced AI systems with very little human involvement.
The company stressed that this has not happened yet and may never happen. However, it believes governments, regulators, and technology companies should start preparing now.
“We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for,” Anthropic said.
Claude Is Playing a Bigger Role in AI Development
Anthropic says its Claude AI model has grown far beyond a basic coding assistant.
The company described how AI tools have evolved. At first, engineers wrote code themselves. Later, chatbots helped with small coding tasks. Then came AI coding agents that could edit files and manage projects on their own.
Today, AI agents can write and run code, complete tasks independently, and even assign work to other AI agents. According to Anthropic, the next step is AI systems that actively help build and train future AI models.
The company revealed that Claude now writes most of the code used to develop Anthropic’s own AI systems. As a result, developers are much more productive. Anthropic says the average engineer now produces about eight times more code per day than in 2024.
Related: Anthropic President Says Rising AI Training Costs Are Driving IPO Plans
AI Capabilities Are Growing Faster
Anthropic says public tests and its internal research show that AI capabilities are improving rapidly. The company claims the length and complexity of tasks AI can reliably complete is doubling about every four months. Previously, this growth rate was closer to every seven months.
Anthropic pointed to the progress of its Claude models as an example. In March 2024, Claude Opus 3 could handle software engineering tasks that would take a human about four minutes. One year later, Claude Sonnet 3.7 could reportedly complete tasks requiring around 90 minutes of human work.
By 2026, Claude Opus 4.6 was reportedly capable of handling tasks equivalent to roughly 12 hours of human effort.
Claude’s research abilities have also improved. In one internal project, Claude-powered agents solved nearly all of a major research challenge.
Anthropic also reported that Claude’s success rate on difficult coding tasks reached 76% in May 2026, a major improvement compared with six months earlier.
Three Possible Futures for AI
Anthropic outlined three possible paths for AI development. The first is a slowdown, where progress becomes harder to achieve over time.
The second is a future where AI continues to boost productivity while humans remain in control of important decisions and oversight.
The third and most dramatic scenario is one where AI systems begin creating their own successors, leading to much faster progress.
Based on current evidence, Anthropic believes the second scenario is the most likely. However, the company says there is still uncertainty about how quickly AI capabilities could advance.
Human Oversight Could Become the Biggest Challenge
Anthropic says humans still have important strengths, including judgment, strategic thinking, and deciding which problems are worth solving. However, the company argues that the biggest challenge may soon be human oversight rather than AI development itself.
As AI systems become more capable, it may become increasingly difficult for people to review, verify, and safely monitor their work.
Because of this, Anthropic suggests governments and leading AI companies should have plans in place to temporarily slow the development of the most advanced AI systems if progress begins moving faster than society can safely manage.
The company noted that any meaningful slowdown would likely require international cooperation. If only one company pauses development, competitors could continue advancing, reducing the effectiveness of the safety measure.
For now, Anthropic says recursive self-improvement remains a possibility, not a certainty. But as Claude plays a growing role in building future AI systems, the company believes discussions about AI governance, oversight, and safety should begin sooner rather than later.
Related: Is AI Killing Bitcoin? Why Capital Is Rotating to AI Tech Now
Disclaimer: The information presented in this article is for informational and educational purposes only. The article does not constitute financial advice or advice of any kind. Coin Edition is not responsible for any losses incurred as a result of the utilization of content, products, or services mentioned. Readers are advised to exercise caution before taking any action related to the company.
