Claude already writes more than 90% of Anthropic's code and multiplies the performance of his engineers by 45

Artificial intelligence is radically transforming the tech industry, and Anthropic has just revealed one of the most compelling examples of that revolution The American company confirmed that its model Claude wrote more than 90% of the contributions incorporated into its code during September 2026, a figure that demonstrates how much these tools have begun to replace programming tasks that previously depended directly on engineers

The data was shared by the company led by Dario Amodei in an update to the S-1/A form submitted to the U.S. Securities and Exchange Commission (SEC) as part of preparations for its IPO

The productivity of its engineering teams also experienced extraordinary growth According to Anthropic, the average daily volume of code incorporated per engineer during the three months ending September 30, 2026, was more than 45 times higher than the corresponding daily average for 2024

The phenomenon represents a significant advance for software development, but it also raises questions about the limits of a technology that is already directly involved in creating its own future versions

Claude already writes more than 90% of Anthropic's code

The results shared by the company show a profound change in the way its engineers develop tech products

Claude has evolved from functioning as an assistant capable of suggesting code snippets to becoming a tool that writes, executes, and reviews programs, and can work for extended periods on complex tasks

This evolution has allowed Anthropic professionals to focus more on directing and supervising the work generated by artificial intelligence, rather than manually writing each line of code

According to official data, more than 90% of the contributions incorporated into the company's codebase during September were written by Claude

The growth is also reflected in the system's performance against complex engineering problems that are open and without fully defined specifications

Anthropic reported that Claude achieved a success rate close to 90% on these tasks during September, according to its internal assessments, after improving by approximately 50 percentage points in nine months

However, the company clarified that the fact that Claude generates most of the code does not mean that it designs future generations of artificial intelligence completely autonomously

For now, research and development of new models continue under human guidance

Anthropic warns about the risks of artificial intelligence accelerating its own development

The technological advance also raises concerns within the company

In its presentation to the SEC, Anthropic acknowledged that using artificial intelligence to develop increasingly powerful artificial intelligence systems could pose a considerable risk to safety

The concern relates to the possibility that the pace of technological advancements could accelerate sharply, surpassing researchers' ability to adequately understand and supervise the new systems

This phenomenon is linked to the concept of recursive self-improvement, a process by which an artificial intelligence contributes to the development of progressively more advanced versions of itself

The company uses the expression closing the loop to describe how the capabilities of one generation of models can accelerate the construction of the next

In this context, Anthropic recognizes that the speed of technological progress could become a challenge for the very control and oversight mechanisms

The company acknowledges that it still needs to maintain human control

One of the most relevant aspects of the document presented by Anthropic is the acknowledgment of the current limits of technological oversight

The company states that it does not foresee being able to use Claude safely to design and build successive generations of cutting-edge artificial intelligence models without significant human intervention, even if that possibility were to become technically feasible

For this reason, Anthropic is studying modifications to its Responsible Scaling Policy to ensure that people continue to control the development of the most advanced systems

The company also warns that there are difficulties in measuring the actual speed of progress

The indicators used to assess the capabilities of its models may provide delayed information, meaning that a significant acceleration could only be identified after it has occurred

In fact, Anthropic acknowledged that its estimates about future generations of models indicate that the threshold defined in its policy as drastically accelerated progress by artificial intelligence could soon be reached

However, the company clarified that none of its current models have been classified as exceeding that threshold

This distinction is fundamental: the risks highlighted by Anthropic relate to the possible capabilities of future systems and do not constitute confirmation that Claude has acquired the ability to develop completely on its own

Anthropic imposes new restrictions on Claude Code to protect intellectual property

As it advances in the development of more autonomous systems, Anthropic has also begun to set specific limits on certain uses of its tools

On October 30, 2026, the company confirmed a modification in the instructions of Claude Code, its programming assistant, to prevent users from employing it to reproduce existing video games

The new rule prohibits generating recreations of specific titles developed by third parties, even when those projects are presented under different names

The restriction began to be detected around October 27 by a user on the social network X, who identified changes in the system's behavior

Subsequently, Addy Osmani, a member of Anthropic's technical team, publicly confirmed that the company had added two instructions to the operation of Claude Code with the Opus 5.5 model

As he explained, the goal is to prevent the tool from being used to reproduce others' video games in a way that infringes intellectual property rights

The decision comes amid growing concern within the video game industry about how easily certain artificial intelligence tools allow for the imitation of existing projects

Claude Code will allow the creation of original video games, but not the copying of third-party titles

Anthropic clarified that the modification is not intended to prevent the development of video games through artificial intelligence

On the contrary, users will be able to continue creating original games, projects belonging to specific genres, and titles developed from their own designs and materials

The restriction specifically targets the reconstruction of existing video games belonging to other developers

Osmani explained that using Claude to infringe intellectual property rights contradicts the company's policies and that the new instruction aims to prevent such practices

However, he also acknowledged that the limitations may cause inconveniences when the system incorrectly rejects legitimate requests

For this reason, he urged users to report cases where Claude refuses to work with materials they own rights to

The company assured that it will continue adjusting the instructions to find a balance between protecting intellectual property and allowing developers to freely use their own creations

This issue presents a significant challenge: preventing artificial intelligence from facilitating the appropriation of others' work without establishing restrictions that ultimately harm innovation and independent development

The artificial intelligence revolution poses a challenge for innovation and technological freedom

The results presented by Anthropic demonstrate the potential of artificial intelligence to transform business productivity

That a system can generate more than 90% of the contributions to a tech company's code demonstrates the magnitude of the change the software industry is undergoing

The growth of more than 45 times in the average daily volume of code incorporated per engineer also reflects how these tools can significantly expand the production capacity of human teams

However, the increase in generated code does not necessarily imply an equivalent improvement in its quality, nor does it allow for the conclusion that the intervention of specialists has ceased to be necessary

Anthropic's experience shows that technological innovation, intellectual property protection, and human responsibility must advance together

On one hand, artificial intelligence tools offer extraordinary opportunities to accelerate the development of new products and reduce barriers to entry in programming

On the other hand, their increasing autonomy demands the establishment of mechanisms that allow for understanding, controlling, and correcting their decisions

The challenge will be to harness the potential of artificial intelligence without losing human control over its development or unnecessarily limiting the freedom of innovation

Anthropic has already demonstrated that Claude can become a central piece of its own technological evolution Now, the company faces the task of ensuring that this acceleration continues to be accompanied by effective human oversight