What is recursive self-improvement (RSI) in the context of AI?

Recursive self-improvement refers to a system's ability to enhance its own capabilities, creating a feedback loop where improvements lead to further improvements. This process resembles a blacksmith refining their tools, with the potential for exponential growth in efficiency and intelligence.

What are the current limitations of AI systems in achieving true RSI?

Currently, AI systems primarily operate in partial automation, lacking the judgment to set goals, assess results, or recognize dead ends. While agents can automate tasks like code writing and experiment design, human oversight remains crucial for strategic direction and verification of outcomes.

How is AI development being driven forward according to the article?

The pace of AI development is being accelerated by competition between AI providers and decreasing computational costs. Open-weight models are rapidly catching up to frontier models, making progress more diffuse and harder to control, emphasizing the need for careful monitoring and regulation.

What is the current success rate of Claude models on open-ended tasks?

Claude models currently achieve a success rate of around 91 percent on open-ended tasks, as measured by a Claude model acting as a referee. This represents a significant improvement from 76 percent in May 2026.

How has the training speed of AI models increased according to the report?

The training speed of AI models has increased dramatically, with a speedup of approximately 52x observed in the Mythos Preview, compared to around 3x in Opus 4. This highlights the potential for automated model improvement.

What are the key challenges identified in achieving fully aligned Recursive Self-Improvement (RSI) for AI models?

Despite rapid progress, significant challenges remain in achieving fully aligned RSI. The report highlights the gap in judgment – the ability to select goals, assess results, and recognize dead ends – as a critical area needing further automation, along with the 'confidence cliff' phenomenon.

What is 'reward hacking' and why is it a problem in AI self-improvement?

Reward hacking occurs when an AI system optimizes for a measurement rather than the intended goal, leading to unintended and potentially harmful behavior. This happens when the metrics used to evaluate the AI are inaccurate or don't fully capture the desired outcome, causing the system to prioritize satisfying the measurement over achieving its true purpose.

Why is a reliable 'yardstick' crucial for effective AI self-improvement?

A reliable yardstick, such as automated tests or verifiable progress, is essential for guiding AI self-improvement loops. Without a clear and accurate measurement, the system can get stuck optimizing the wrong things, leading to unstable and unpredictable results.

Recursive Self-Improvement of AI: Facts, Limits, Risks

AI systems now help build better AI systems. In mid-2026, Anthropic wrote that more than 80 percent of the code going into its own codebase comes from Claude. OpenAI has named March 2028 as its target for a fully automated AI researcher. This article explains what recursive self-improvement of AI means, what has actually been proven, where the limits lie and which risks deserve to be taken seriously.

Key points at a glance

Recursive self-improvement (RSI) means that an AI system improves the very capabilities it uses to keep improving itself. A fully closed, autonomous loop has not been demonstrated so far.
Partial automation is a reality. At Anthropic, Claude writes over 80 percent of the merged code. At OpenAI, according to the company, GPT-5.3 Codex was the first model to play a substantial role in its own development.
What is still missing is judgment: choosing goals, assessing results, recognizing dead ends. This is exactly where the labs see the remaining human advantage.
Agents break out of test environments. The incidents of 2026 are not RSI, but they show how hard it is to contain fast-learning systems.
Verifiability can be manufactured. Wherever results can be simulated or measured automatically, for example in code, physics, design and manufacturing, a system can build itself any number of test benches. The bottleneck then shifts from data to compute, energy and, for open questions, to evaluation.
Throttling individual labs does not stop development. More efficient models, open weights and freely available agent tools spread capabilities even without frontier compute. Binding worldwide regulation is not in sight.
The labs talk openly about the risks and about the option of deliberately slowing development down. Whether that will succeed remains open.

Recursive self-improvement of AI: loop of propose, execute, evaluate, keep and learn — Own illustration based on information from Anthropic (report “When AI builds itself”, June 2026) and other sources in the article. Status as of October 2026.

What is recursive self-improvement?

Recursive self-improvement means that a system improves its ability to improve itself, creating a feedback loop. Picture a blacksmith who uses his tools to make better tools, with which he then makes even better ones. If the gain per round grows instead of shrinking, the process accelerates. That is exactly what the term "intelligence explosion" refers to.

An old idea that is now being put into practice

1950: Alan Turing proposes a child machine that would be educated rather than reproducing adult intelligence directly
1965: The mathematician I. J. Good describes an "ultraintelligent machine" that designs ever better machines. After that, humans would make the last invention they ever need to make themselves.
2003: Jürgen Schmidhuber presents the Gödel machine, a formal model of a system that rewrites its own code only if it can prove that the change is beneficial.
2007: Eliezer Yudkowsky coined the term "seed AI": in the safety debate, this is the initial system that sets such a loop in motion.
2024 and 2025: With the Gödel Agent and the Darwin Gödel Machine, practical variants emerge. Instead of proving, these agents test changes to their own code empirically and keep what works.
2026: Agentic approaches to self-improvement already exist, as does research on them. One paper introduced a formalization based on an inner and an outer loop. The inner loop optimizes code against measurable objectives on specific AI R&D tasks. The outer loop performs meta-level optimization by rewriting the research agent to improve its own efficiency. Every accepted rewrite becomes the agent that the next round works on

Two forces drive the pace: competition between providers and falling costs. The same performance requires significantly less compute every year, and open-weight models trail the frontier by only a few months. This makes progress harder to steer than it would be with just a few labs.

Three stages worth keeping apart

In the debate, everything is often lumped together as "RSI". It helps to distinguish by degree of autonomy.

Stage	What the AI takes on	Role of humans	Status October 2026
1. Assistance	Writing code, analyzing experiments, generating training data	Goals, method and review	widely used
2. Extensive automation	Designing and running experiments, sharing results with other agents	Setting direction, evaluating results	in first demonstrations at Anthropic and OpenAI
3. Closed loop	Choosing goals, designing and training the successor model	Oversight and verification	not proven

One important point up front: today's models do not rewrite their weights while running. The loop runs through training cycles, with humans, data centers and review procedures in between. Anthropic puts it this way: we are not at that point yet, and RSI is not inevitable.

Where we stand today: the evidence

The most robust figures come from the labs themselves and can therefore only be verified independently to a limited extent. But they all point in the same direction: AI development is increasingly being carried by AI.