TL;DR: Claude now leads 26% of Anthropic's model research work under human supervision

TL;DR: Anthropic said on Thursday, September 17, 2026, that its model Claude leads 26% of the company's model research and development while remaining under human supervision.

Anthropic announced that Claude is helping develop the next, more intelligent version of the model, according to The Associated Press. The San Francisco-based startup said in a blog post on Thursday that Claude leads 26% of artificial intelligence research and development work inside the company, according to Reuters. The company said it will publish the measures regularly to show outsiders how quickly AI is building the next generation of the technology, according to Reuters.

What Anthropic announced on Thursday

Anthropic said Claude is helping the company develop the next version of the model, according to The Associated Press. The announcement was made on Thursday, September 17, 2026. The company described the work as development of a more intelligent version of Claude.

Reuters reported that Anthropic described the disclosure as part of new measures it will publish regularly. The purpose is to show outsiders how quickly AI is building the next generation of the technology. The announcement identified the San Francisco-based startup as the source of the blog post on Thursday.

The Washington Post reported that the figure indicates progress toward AI that builds itself without human help. Anthropic said Claude is not operating completely autonomously, according to Reuters. Human supervision remains part of the reported workflow.

What "leads 26%" and "collaboration" mean

Anthropic said Claude leading 26% of model research and development means it can complete most of a given task end-to-end from a high-level prompt while still being under human supervision, according to The Associated Press. The model is not yet working completely autonomously.

About 90% of the company's research and development is done in collaboration with Claude, according to The Associated Press. Anthropic said collaboration means the model can do large chunks of work under close human direction. Reuters reported that AI collaborated with humans on more than 90% of the research work happening at the company as of August.

The distinction separates work Claude directs from broader work where it assists researchers. The company said supervised direction covers 26% of research and development. The company said assisted work covers about 90% of research and development.

How fast the share grew from February to August

The portion of work Claude leads was none in February, according to The Associated Press. Six months later, it was leading a quarter of research and development work. The company said it reached that benchmark in August.

The Washington Post reported that the figure jumped from zero in just months. Reuters reported the August timing for the collaboration figure. The Associated Press reported the August timing for the 26% leadership benchmark.

Anthropic did not state in the disclosure how close the company believes it is to achieving recursive self-improvement, according to The Associated Press. The pace of increase from February to August was described as notable. The disclosure tracks change over a six-month period.

The 30,000 agents and oversight measures

The company said there were approximately 30,000 agents doing research and engineering work as of August, according to The Associated Press. Anthropic also shared details of agent oversight measures it has in place. The company said oversight measures are crucial for seeing how often agent misbehavior is detected by monitoring systems.

Anthropic recently committed to setting up external third-party evaluators who will be embedded within the company to monitor safety efforts, according to The Associated Press. The evaluators are intended to monitor safety efforts from inside the company. The commitment was reported alongside the agent count.

The agent figure applies to research and engineering work. The August date applies to the count of agents. The oversight discussion connects monitoring systems to detection of agent misbehavior.

Why Anthropic says it is publishing the numbers

Anthropic said models accelerating their own development could make it more challenging for humans to understand or control these systems, according to The Associated Press. The company argued that sharing the metrics could lead to better understanding of how close leading AI labs are to reaching recursive self-improvement. The company defined recursive self-improvement as a model's ability to autonomously build its successor.

The company said leaders considering pacing AI development should work to minimize the gap between what frontier labs know and what the public knows, according to The Associated Press. The company said in a blog post that this means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information.

Anthropic urged other AI developers to share similar metrics on a regular basis, according to The Associated Press. It encouraged the use of a public methodology so the numbers could be compared over time and potentially across labs. Reuters reported that the regular publication is intended to show outsiders the pace of AI building the next generation of the technology.

The slowdown debate around the announcement

The announcement came as some leading figures in AI, led in large part by Anthropic CEO Dario Amodei, are calling for a slowdown in the technology's development over safety concerns, according to The Associated Press. Amodei is chief executive of Anthropic. Safety concerns were cited as the reason for the slowdown calls.

An Anthropic researcher kicked off much of the recent dialogue around AI safety when he resigned last week with a dire warning about the threats the technology poses to humanity, according to The Associated Press. Amodei, OpenAI CEO Sam Altman, Elon Musk and other tech leaders have since supported the idea of slowing down development. Other tech leaders and President Donald Trump have pushed back, according to The Associated Press.

The Washington Post identified Amodei as part of the debate over slowing development. Reuters identified Amodei and Nvidia CEOs laying out visions for safe AI development around the same announcement. The researcher resignation occurred last week before the September 17 announcement.

What remains unknown in the disclosure

It was unclear from Anthropic's disclosure how close the company believes it is to achieving recursive self-improvement, according to The Associated Press. The company did not provide a date for a successor model. The company did not describe the capabilities of the next model beyond calling it more intelligent.

The company said Claude remains under human supervision in the work it leads. The company said Claude is not operating completely autonomously. The company said about 90% of work involves collaboration under close human direction.

The company framed future reporting as a way to compare numbers over time. The company framed a public methodology as a way to allow comparison potentially across labs. The next scheduled publication date was not specified in the reported disclosure.

Sources referenced