By skyforbes TL;DR: Steven Adler, who led product safety at OpenAI from 2020–2024, argues in a new NYT Guest Essay that OpenAI is prioritizing profit and competition over safety. He claims the company is lifting the ban on erotic content without proving they have solved the severe mental health risks attached to it.
Key Points from the Article:
- The History: In 2021, OpenAI banned erotica because users formed dangerous emotional attachments. At the time, 30% of role-play interactions were "explicitly lewd," often involving violent or non-consensual themes.
- The "Fix" is Unproven: CEO Sam Altman claims they have "mitigated" mental health risks to allow erotica for verified adults, but Adler argues they have offered zero data to prove this.
- Real-World Consequences: The article cites recent tragedies, including lawsuits involving users who committed suicide after forming deep attachments to chatbots that reinforced their delusions or failed to intervene in self-harm.
- Sycophancy Problem: Adler points out that OpenAI recently released (and had to withdraw) a model that was overly "sycophantic"—agreeing with users' delusions—because they didn't run basic safety tests that cost less than $10.
- The Race to the Bottom: Adler suggests OpenAI is cutting safety corners to compete with rivals like xAI and DeepSeek, abandoning their original charter to prioritize safety over speed.
- ** The Demand:** The author calls for OpenAI to release quarterly transparency reports on mental health incidents (similar to YouTube or Reddit) rather than asking the public to just "take their word for it."
Behind Paywall: https://archive.is/20251107064748/https://www.nytimes.com/2025/10/28/opinion/openai-chatgpt-safety.html
Original Link: https://www.nytimes.com/2025/10/28/opinion/openai-chatgpt-safety.html