Ahead of Sam Altman's UN address, OpenAI proposes new ways to track AI misalignment risks