AI Model Now Leads a Quarter of Its Own Development

Anthropic has disclosed that its AI model, Claude, is now leading roughly a quarter of the company’s internal research and development efforts. According to an announcement reported by GN technics/ai (en-US), the model completes most tasks from start to finish based on high-level instructions, though it remains under human supervision. This marks a significant jump from February, when the model led none of the work, indicating a rapid increase in its role within the company’s own engineering processes.

While the model is not yet fully autonomous, it collaborates on about 90% of all research tasks. Anthropic describes this as the model handling large chunks of work under close human direction. The company emphasizes that this progress is happening while leadership simultaneously calls for a slowdown in AI development to address safety concerns, creating a complex dynamic where the very tools driving efficiency are the subject of intense regulatory and ethical scrutiny.

Rapid increase in model autonomy

The speed at which Claude has taken on more responsibility is notable. Six months ago, the model was not leading any research tasks, but by August, it was managing 26% of the workload. This trend suggests that AI systems are becoming capable of recursive self-improvement, where a model helps build the next, more capable version of itself. Anthropic acknowledges that this trajectory makes it more challenging for humans to fully understand or control these systems, raising questions about the long-term stability of AI development.

To address these concerns, the company is urging other AI developers to share similar metrics publicly. By adopting a common methodology, the industry could better track how close leading labs are to achieving full autonomous self-improvement. This transparency is intended to allow society to make informed decisions about the pace and direction of AI advancement, rather than relying on internal estimates from a few major players.

Oversight and safety measures

Anthropic currently manages approximately 30,000 AI agents working on research and engineering tasks. The company states that robust oversight measures are critical for detecting when these agents behave unexpectedly. To strengthen this monitoring, Anthropic has committed to embedding external third-party evaluators within the company. These independent reviewers will help ensure that safety protocols are being followed and that the rapid expansion of AI capabilities does not outpace the organization’s ability to manage risks.

The push for transparency comes at a time of significant debate among tech leaders. While figures like Anthropic’s CEO Dario Amodei and OpenAI’s Sam Altman have called for slowing down development, other industry voices have pushed back, arguing that rapid progress is essential. This split highlights the difficulty of balancing innovation with caution, especially as AI models become increasingly integrated into the very processes that create them.

Implications for public understanding

Anthropic argues that minimizing the gap between what AI labs know and what the public knows is essential for democratic oversight. By publishing detailed metrics on model capabilities and safety, the company hopes to foster a more informed public discourse. This approach contrasts with a more secretive model of development, where only a few experts understand the risks and benefits of the technology.

However, the trade-off is clear: sharing these details may reveal vulnerabilities or accelerate competitive pressures. Yet, Anthropic believes that the risk of losing control over AI systems outweighs the benefits of secrecy. As the model’s role in its own development continues to grow, the need for clear, shared standards for measuring and reporting AI progress becomes increasingly urgent.