OpenAI reported new safety incidents and set a disclosure plan, according to Bloomberg. Reporter Shirin Ghaffary said the disclosures highlight unresolved AI alignment risks as models become more powerful.