OpenAI hits self-improving AI milestone, warns of control risks
openai reports ai 'research interns' now handle scoped tasks, but chief scientist warns no lab has solved control of these systems well enough.
openai has announced it has reached a self-set milestone on the path toward self-improving ai, with ai agents now driving its own model development. the company reports that its 'automated research intern' can handle clearly scoped research tasks under human guidance. according to openai's internal metrics, agents now do three times the human workday, with the median researcher burning over $600 a day in inference. the company aims to build a fully automated ai researcher by march 2028.