Steven Adler Flags Irreversible AI Action Risks

Steven Adler, an independent AI safety researcher formerly at OpenAI, posted a reply in a Clear-Eyed AI comment thread on supervising AI. He noted a difference in concern levels with another poster, stating he imagines a higher chance of AI doing something that cannot easily be come back from. In a follow-up visible reply, Adler addressed self-exfiltration likelihood and wondered whether an AI company CEO would agree. The posts appear in the thread linked from clear-eyed.ai.