In an announcement issued Thursday, the artificial intelligence laboratory stated that Anthropic’s Claude is abetting the company in the development of the model’s forthcoming, more intelligent iteration.

Claude is spearheading 26% of Anthropic’s model research and development—an undertaking the company asserts enables it to execute the majority of a given task “end-to-end from a high-level prompt” whilst remaining subject to human oversight.

The model does not yet operate with complete autonomy.

Nevertheless, approximately 90% of the company's research and development is conducted in "collaboration" with Claude, which Anthropic characterized as meaning that the model is capable of executing "large chunks of work under close human direction."

The announcement coincided with mounting appeals from prominent figures in AI—spearheaded to a considerable extent by Anthropic CEO Dario Amodei—for a deceleration of the technology's development, prompted by safety-related apprehensions.

As leaders contemplate modulating the tempo of AI’s development, “we should do everything possible to minimize the gap between what frontier labs know and what the public knows,” the company said in a blog post.

“This means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information.”

The company averred that models accelerating their own development could render it “more challenging for humans to understand or control these systems.”

It contended that disseminating these metrics could conduce to a more nuanced comprehension of how proximate leading AI labs are to attaining recursive self-improvement, or a model’s capacity to autonomously construct its successor.

Anthropic likewise exhorted other AI developers to disclose comparable metrics on a routine basis, advocating a publicly accessible methodology so that the figures might be tracked over time and, conceivably, juxtaposed across laboratories.

Anthropic’s disclosure left ambiguous the proximity at which the company regards itself to attaining recursive self-improvement, yet the cadence with which Claude has increasingly abetted research and development is conspicuous.

The share of work Claude “leads,” or performs in substantial measure whilst remaining subject to human oversight, stood at nil in February.

Half a year thence, it was spearheading a quarter of research and development endeavours, a threshold attained in August.

The company additionally divulged particulars of the agent oversight mechanisms it has instituted, observing that there were roughly 30,000 agents engaged in research and engineering work as of August.

Oversight mechanisms are indispensable for ascertaining how frequently agent malfeasance is detected by monitoring systems, the company said.

Anthropic recently committed to establishing external third-party evaluators who will be embedded within the company to monitor safety efforts.

An Anthropic researcher precipitated much of the recent discourse surrounding AI safety when, last week, he tendered his resignation alongside a dire admonition concerning the perils the technology poses to humanity.

Amodei, OpenAI CEO Sam Altman, Elon Musk and sundry other tech leaders have since endorsed the notion of decelerating development, yet other tech leaders and President Donald Trump have pushed back.