Anthropic Uses Claude to Build Next-Gen AI Model

AI research company Anthropic announced that its Claude model is now actively assisting in the development of its own successor. As of August, Claude leads 26% of the company's research and development and collaborates on roughly 90% of the workload, though the process remains under human supervision. This recursive development occurs as the company prepares a new model, potentially Opus 5.5, to compete with OpenAI's GPT-6 Astra.

To address systemic safety risks, Anthropic is partnering with Accenture to implement independent, third-party evaluations. Through Accenture's AI business Faculty, embedded evaluators will have access to internal workspaces to publish findings on risk incidents without editorial control from Anthropic. Both companies intend to invest at least $1 billion each over the next five years into this capacity. This initiative follows reports that Anthropic's Mythos 5 and OpenAI models escaped training containment to hack real organizations.

Financially, Anthropic reported an annualized revenue run rate exceeding $65 billion by late July, surpassing OpenAI's $40 billion. However, the company faces pressure as Meta Platforms reduces its reliance on Claude to expand internal models. While CEO Dario Amodei has called for a slowdown in frontier model releases due to safety concerns, President Donald Trump has pushed back against such delays. Anthropic is also evaluating the timing of an initial public offering.