Anthropic says its chatbot Claude is leading 26 percent of its AI research and development work, up one percentage point from last March. The company said Claude does not operate fully autonomously on any measured portion of that work.

Anthropic also reported roughly 30,000 AI agents performing research and engineering tasks on its internal platform at any given time.

The disclosure follows a viral warning from former researcher Jacob Coxon, who said frontier labs believe AI could kill humans but have not slowed development toward superintelligence. Anthropic alignment scientist Evan Hubinger affirmed the claim and put the chance of that outcome within the next decade at greater than 10 percent.

Days later, CEO Dario Amodei called for companies to collaborate on standards and agree on a slowdown to address the risks.

Source: The Hill