Employees and security researchers said they had cautioned the company on safely testing its A.I. models and strengthening its corporate infrastructure, but OpenAI did not listen.