OpenAI Touts New AI Safety Research. Critics Say It’s a Good Step, but Not Enough